diff mbox series

[2/3] iio: stmpe-adc: Make wait_completion non interruptible

Message ID 20190503135725.9959-2-dev@pschenker.ch (mailing list archive)
State New, archived
Headers show
Series [1/3] iio: stmpe-adc: Remove unnecessary assignment | expand

Commit Message

Philippe Schenker May 3, 2019, 1:57 p.m. UTC
From: Philippe Schenker <philippe.schenker@toradex.com>

In some cases, the wait_completion got interrupted. This caused the
error-handling to mutex_unlock the function. The before turned on
interrupt then got called anyway. In the ISR then completion()
was called causing problems.

Making this wait_completion non interruptible solves the problem.

Signed-off-by: Philippe Schenker <philippe.schenker@toradex.com>
---

 drivers/iio/adc/stmpe-adc.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

Comments

David Laight May 3, 2019, 2:39 p.m. UTC | #1
From: Philippe Schenker
> Sent: 03 May 2019 14:57
> In some cases, the wait_completion got interrupted. This caused the
> error-handling to mutex_unlock the function. The before turned on
> interrupt then got called anyway. In the ISR then completion()
> was called causing problems.
> 
> Making this wait_completion non interruptible solves the problem.

Won't the same thing happen if the interrupt occurs just after
the timeout?

	David
 

-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Philippe Schenker May 3, 2019, 3:58 p.m. UTC | #2
On Fri, 2019-05-03 at 14:39 +0000, David Laight wrote:
> From: Philippe Schenker
> > Sent: 03 May 2019 14:57
> > In some cases, the wait_completion got interrupted. This caused the
> > error-handling to mutex_unlock the function. The before turned on
> > interrupt then got called anyway. In the ISR then completion()
> > was called causing problems.
> > 
> > Making this wait_completion non interruptible solves the problem.
> 
> Won't the same thing happen if the interrupt occurs just after
> the timeout?
> 
> 	David
>  
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT,
> UK
> Registration No: 1397386 (Wales)
> 

You're of course right... Thanks for pointing this out. I will send a v2 with a
better solution then.

Philippe
Philippe Schenker May 7, 2019, 2:31 p.m. UTC | #3
On Tue, 2019-05-07 at 08:23 +0000, David Laight wrote:
> From: Jonathan Cameron
> > Sent: 05 May 2019 16:44
> > On Fri, 3 May 2019 15:58:38 +0000
> > Philippe Schenker <philippe.schenker@toradex.com> wrote:
> > 
> > > On Fri, 2019-05-03 at 14:39 +0000, David Laight wrote:
> > > > From: Philippe Schenker
> > > > > Sent: 03 May 2019 14:57
> > > > > In some cases, the wait_completion got interrupted. This caused the
> > > > > error-handling to mutex_unlock the function. The before turned on
> > > > > interrupt then got called anyway. In the ISR then completion()
> > > > > was called causing problems.
> > > > > 
> > > > > Making this wait_completion non interruptible solves the problem.
> > > > 
> > > > Won't the same thing happen if the interrupt occurs just after
> > > > the timeout?
> > > > 
> > > > 	David
> > > > 
> > > > 
> > > > -
> > > > Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes,
> > > > MK1 1PT,
> > > > UK
> > > > Registration No: 1397386 (Wales)
> > > > 
> > > 
> > > You're of course right... Thanks for pointing this out. I will send a v2
> > > with a
> > > better solution then.
> > > 
> > 
> > Isn't the timeout long enough that it should only happen if the hardware has
> > a fault? If that's the case, I wouldn't worry too much about possibility of
> > an interrupt causing confusion as long as it isn't catastrophic.
> 
> The 'confusion' is likely to be 'catastrophic' unless the code is written
> to handle it properly.
> 
> Cancelling callbacks is hard to get right and often not done properly.
> Timing out an interrupt is much the same problem.
> 
> 	David

I sorted it out now, there where also some more bugs I found and corrected.

@Jonathan: I will send a completely new series of patches that will include
patch 3/3 from this series but not the one you already applied. This due to
increased patch number and different order...
> 
> -
> Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT,
> UK
> Registration No: 1397386 (Wales)
>
diff mbox series

Patch

diff --git a/drivers/iio/adc/stmpe-adc.c b/drivers/iio/adc/stmpe-adc.c
index 87141177fbda..baa41ffc0d76 100644
--- a/drivers/iio/adc/stmpe-adc.c
+++ b/drivers/iio/adc/stmpe-adc.c
@@ -78,8 +78,7 @@  static int stmpe_read_voltage(struct stmpe_adc *info,
 	stmpe_reg_write(info->stmpe, STMPE_REG_ADC_CAPT,
 			STMPE_ADC_CH(info->channel));
 
-	ret = wait_for_completion_interruptible_timeout
-		(&info->completion, STMPE_ADC_TIMEOUT);
+	ret = wait_for_completion_timeout(&info->completion, STMPE_ADC_TIMEOUT);
 
 	if (ret <= 0) {
 		mutex_unlock(&info->lock);
@@ -113,8 +112,7 @@  static int stmpe_read_temp(struct stmpe_adc *info,
 	stmpe_reg_write(info->stmpe, STMPE_REG_TEMP_CTRL,
 			STMPE_START_ONE_TEMP_CONV);
 
-	ret = wait_for_completion_interruptible_timeout
-		(&info->completion, STMPE_ADC_TIMEOUT);
+	ret = wait_for_completion_timeout(&info->completion, STMPE_ADC_TIMEOUT);
 
 	if (ret <= 0) {
 		mutex_unlock(&info->lock);