From: Bartosz Golaszewski <[email protected]>
i2c_davinci_cpufreq_transition() is implemented in a way that will
block if it ever gets called while no transfer is in progress.
Not only that, but reinit_completion() is never called for xfr_complete.
Use the fact that cpufreq uses an srcu_notifier (running in process
context) for transitions and that the bus_lock is taken during the call
to master_xfer() and simplify the code by removing the transfer
completion entirely and protecting i2c_davinci_cpufreq_transition()
with i2c_lock/unlock_adapter().
Reported-by: David Lechner <[email protected]>
Signed-off-by: Bartosz Golaszewski <[email protected]>
---
drivers/i2c/busses/i2c-davinci.c | 12 ++++--------
1 file changed, 4 insertions(+), 8 deletions(-)
diff --git a/drivers/i2c/busses/i2c-davinci.c b/drivers/i2c/busses/i2c-davinci.c
index 2ead9b9eebb7..53ebf0dec1ad 100644
--- a/drivers/i2c/busses/i2c-davinci.c
+++ b/drivers/i2c/busses/i2c-davinci.c
@@ -139,7 +139,6 @@ struct davinci_i2c_dev {
u8 terminate;
struct i2c_adapter adapter;
#ifdef CONFIG_CPU_FREQ
- struct completion xfr_complete;
struct notifier_block freq_transition;
#endif
struct davinci_i2c_platform_data *pdata;
@@ -567,9 +566,6 @@ i2c_davinci_xfer(struct i2c_adapter *adap, struct i2c_msg msgs[], int num)
}
ret = num;
-#ifdef CONFIG_CPU_FREQ
- complete(&dev->xfr_complete);
-#endif
out:
pm_runtime_mark_last_busy(dev->dev);
@@ -717,13 +713,15 @@ static int i2c_davinci_cpufreq_transition(struct notifier_block *nb,
struct davinci_i2c_dev *dev;
dev = container_of(nb, struct davinci_i2c_dev, freq_transition);
+
+ i2c_lock_adapter(&dev->adapter);
if (val == CPUFREQ_PRECHANGE) {
- wait_for_completion(&dev->xfr_complete);
davinci_i2c_reset_ctrl(dev, 0);
} else if (val == CPUFREQ_POSTCHANGE) {
i2c_davinci_calc_clk_dividers(dev);
davinci_i2c_reset_ctrl(dev, 1);
}
+ i2c_unlock_adapter(&dev->adapter);
return 0;
}
@@ -789,9 +787,7 @@ static int davinci_i2c_probe(struct platform_device *pdev)
}
init_completion(&dev->cmd_complete);
-#ifdef CONFIG_CPU_FREQ
- init_completion(&dev->xfr_complete);
-#endif
+
dev->dev = &pdev->dev;
dev->irq = irq;
dev->pdata = dev_get_platdata(&pdev->dev);
--
2.15.1
On Monday 22 January 2018 08:12 PM, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <[email protected]>
>
> i2c_davinci_cpufreq_transition() is implemented in a way that will
> block if it ever gets called while no transfer is in progress.
true!
>
> Not only that, but reinit_completion() is never called for xfr_complete.
reinit_completion() use is must only with complete_all(). At least that
bug is not obvious so needs more detail to be added here.
> Use the fact that cpufreq uses an srcu_notifier (running in process
> context) for transitions and that the bus_lock is taken during the call
> to master_xfer() and simplify the code by removing the transfer
> completion entirely and protecting i2c_davinci_cpufreq_transition()
> with i2c_lock/unlock_adapter().
>
> Reported-by: David Lechner <[email protected]>
> Signed-off-by: Bartosz Golaszewski <[email protected]>
I assume you tested this on a board where I2C-controlled PMIC changes
voltage during CPUfreq (DA850 LogicPD EVM should have this hardware
support).
Perhaps also add:
Fixes: 82c0de11b734 ("i2c: davinci: Add cpufreq support")
Apart from that:
Reviewed-by: Sekhar Nori <[email protected]>
Thanks,
Sekhar
On 01/22/2018 08:42 AM, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <[email protected]>
>
> i2c_davinci_cpufreq_transition() is implemented in a way that will
> block if it ever gets called while no transfer is in progress.
>
> Not only that, but reinit_completion() is never called for xfr_complete.
>
> Use the fact that cpufreq uses an srcu_notifier (running in process
> context) for transitions and that the bus_lock is taken during the call
> to master_xfer() and simplify the code by removing the transfer
> completion entirely and protecting i2c_davinci_cpufreq_transition()
> with i2c_lock/unlock_adapter().
>
> Reported-by: David Lechner <[email protected]>
> Signed-off-by: Bartosz Golaszewski <[email protected]>
> ---
Tested that it fixes the lockup when there is no active transfer.
Tested-by: David Lechner <[email protected]>
On Mon, Jan 22, 2018 at 03:42:58PM +0100, Bartosz Golaszewski wrote:
> From: Bartosz Golaszewski <[email protected]>
>
> i2c_davinci_cpufreq_transition() is implemented in a way that will
> block if it ever gets called while no transfer is in progress.
>
> Not only that, but reinit_completion() is never called for xfr_complete.
>
> Use the fact that cpufreq uses an srcu_notifier (running in process
> context) for transitions and that the bus_lock is taken during the call
> to master_xfer() and simplify the code by removing the transfer
> completion entirely and protecting i2c_davinci_cpufreq_transition()
> with i2c_lock/unlock_adapter().
>
> Reported-by: David Lechner <[email protected]>
> Signed-off-by: Bartosz Golaszewski <[email protected]>
Applied to for-next, thanks!