Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1473255 > unrolled thread

A potential bug in drivers/iio/light/opt3001.ko

Started byPavel Andrianov <andrianov@ispras.ru>
First post2016-08-31 12:30 +0200
Last post2016-09-05 22:10 +0200
Articles 4 — 2 participants

Back to article view | Back to linux.kernel

Contents

  A potential bug in drivers/iio/light/opt3001.ko Pavel Andrianov <andrianov@ispras.ru> - 2016-08-31 12:30 +0200
    Re: A potential bug in drivers/iio/light/opt3001.ko Jonathan Cameron <jic23@kernel.org> - 2016-09-03 18:40 +0200
      Re: A potential bug in drivers/iio/light/opt3001.ko Pavel Andrianov <andrianov@ispras.ru> - 2016-09-05 16:20 +0200
        Re: A potential bug in drivers/iio/light/opt3001.ko Jonathan Cameron <jic23@kernel.org> - 2016-09-05 22:10 +0200

#1473255 — A potential bug in drivers/iio/light/opt3001.ko

FromPavel Andrianov <andrianov@ispras.ru>
Date2016-08-31 12:30 +0200
SubjectA potential bug in drivers/iio/light/opt3001.ko
Message-ID<scaWm-1dE-29@gated-at.bofh.it>
Hi!

There is a bug in drivers/iio/light/opt3001.ko. Regard such case:

Thread 1                             Thread 2
-> opt3001_read_raw
   -> mutex_lock(&opt->lock)
   -> opt3001_get_lux()
     ..
     ->i2c_smbus_write_word_swapped()
             Now an interrupt comes
                                      -> opt3001_irq
                                        -> mutex_lock(&opt->lock)

This is a deadlock, as the flag ok_to_ignore_lock has not been set yet.

Regard another case:

Thread 1                             Thread 2
-> opt3001_read_raw
   -> mutex_lock(&opt->lock)
   -> opt3001_get_lux()
     ..
     -> i2c_smbus_write_word_swapped()
     opt->ok_to_ignore_lock = true;
             Now an interrupt comes
                                      -> opt3001_irq
                                        ..
                                        opt->result_ready = true
                                        wake_up()
      opt->result_ready = false;
      wait_event_timeout()

In this case the first thread misses the result and waits until timeout 
expires.

-- 
Pavel Andrianov
Linux Verification Center, ISPRAS
web: http://linuxtesting.org
e-mail: andrianov@ispras.ru

[toc] | [next] | [standalone]

#1475739

FromJonathan Cameron <jic23@kernel.org>
Date2016-09-03 18:40 +0200
Message-ID<sdm94-7Hd-29@gated-at.bofh.it>
In reply to#1473255
On 31/08/16 11:23, Pavel Andrianov wrote:
> Hi!
> 
> There is a bug in drivers/iio/light/opt3001.ko. Regard such case:
> 
> Thread 1                             Thread 2
> -> opt3001_read_raw
>   -> mutex_lock(&opt->lock)
>   -> opt3001_get_lux()
>     ..
>     ->i2c_smbus_write_word_swapped()
>             Now an interrupt comes
>                                      -> opt3001_irq
>                                        -> mutex_lock(&opt->lock)
> 
> This is a deadlock, as the flag ok_to_ignore_lock has not been set yet.
Good find.  Will need reordering to set the ok_to_ignore_lock first.
Whether it ever actually happens will depend on just how long that EOC
interrupt takes to happen.  Still it's a theoretical problem with
a fairly simple fix so let's fix it.
> 
> Regard another case:
> 
> Thread 1                             Thread 2
> -> opt3001_read_raw
>   -> mutex_lock(&opt->lock)
>   -> opt3001_get_lux()
>     ..
>     -> i2c_smbus_write_word_swapped()
>     opt->ok_to_ignore_lock = true;
>             Now an interrupt comes
>                                      -> opt3001_irq
>                                        ..
>                                        opt->result_ready = true
>                                        wake_up()
>      opt->result_ready = false;
>      wait_event_timeout()
> 
> In this case the first thread misses the result and waits until timeout expires.
> 
Agreed - looks like some reordering is needed here as well.

Jonathan

[toc] | [prev] | [next] | [standalone]

#1476484

FromPavel Andrianov <andrianov@ispras.ru>
Date2016-09-05 16:20 +0200
Message-ID<se2UF-4Qu-15@gated-at.bofh.it>
In reply to#1475739
03.09.2016 19:38, Jonathan Cameron пишет:
> On 31/08/16 11:23, Pavel Andrianov wrote:
>> Hi!
>>
>> There is a bug in drivers/iio/light/opt3001.ko. Regard such case:
>>
>> Thread 1                             Thread 2
>> -> opt3001_read_raw
>>   -> mutex_lock(&opt->lock)
>>   -> opt3001_get_lux()
>>     ..
>>     ->i2c_smbus_write_word_swapped()
>>             Now an interrupt comes
>>                                      -> opt3001_irq
>>                                        -> mutex_lock(&opt->lock)
>>
>> This is a deadlock, as the flag ok_to_ignore_lock has not been set yet.
> Good find.  Will need reordering to set the ok_to_ignore_lock first.
> Whether it ever actually happens will depend on just how long that EOC
> interrupt takes to happen.  Still it's a theoretical problem with
> a fairly simple fix so let's fix it.
>>
>> Regard another case:
>>
>> Thread 1                             Thread 2
>> -> opt3001_read_raw
>>   -> mutex_lock(&opt->lock)
>>   -> opt3001_get_lux()
>>     ..
>>     -> i2c_smbus_write_word_swapped()
>>     opt->ok_to_ignore_lock = true;
>>             Now an interrupt comes
>>                                      -> opt3001_irq
>>                                        ..
>>                                        opt->result_ready = true
>>                                        wake_up()
>>      opt->result_ready = false;
>>      wait_event_timeout()
>>
>> In this case the first thread misses the result and waits until timeout expires.
>>
> Agreed - looks like some reordering is needed here as well.
>
> Jonathan
>

In opt3001_get_lux has a comment, that i2c_smbus_write_word_swapped 
(line 246) enables interrupt mechanism. If an interrupt can not arise 
before the function, the assignments to both of flags should be moved 
before i2c_smbus_write_word_swapped and this is the best fix for both of 
issues.
Do you know if my assumption is correct and interrupts are disabled 
before i2c_smbus_write_word_swapped call?
-- 
Pavel Andrianov
Linux Verification Center, ISPRAS
web: http://linuxtesting.org
e-mail: andrianov@ispras.ru

[toc] | [prev] | [next] | [standalone]

#1476928

FromJonathan Cameron <jic23@kernel.org>
Date2016-09-05 22:10 +0200
Message-ID<se8nn-8w0-29@gated-at.bofh.it>
In reply to#1476484
On 05/09/16 15:15, Pavel Andrianov wrote:
> 03.09.2016 19:38, Jonathan Cameron пишет:
>> On 31/08/16 11:23, Pavel Andrianov wrote:
>>> Hi!
>>>
>>> There is a bug in drivers/iio/light/opt3001.ko. Regard such case:
>>>
>>> Thread 1                             Thread 2
>>> -> opt3001_read_raw
>>>   -> mutex_lock(&opt->lock)
>>>   -> opt3001_get_lux()
>>>     ..
>>>     ->i2c_smbus_write_word_swapped()
>>>             Now an interrupt comes
>>>                                      -> opt3001_irq
>>>                                        -> mutex_lock(&opt->lock)
>>>
>>> This is a deadlock, as the flag ok_to_ignore_lock has not been set yet.
>> Good find.  Will need reordering to set the ok_to_ignore_lock first.
>> Whether it ever actually happens will depend on just how long that EOC
>> interrupt takes to happen.  Still it's a theoretical problem with
>> a fairly simple fix so let's fix it.
>>>
>>> Regard another case:
>>>
>>> Thread 1                             Thread 2
>>> -> opt3001_read_raw
>>>   -> mutex_lock(&opt->lock)
>>>   -> opt3001_get_lux()
>>>     ..
>>>     -> i2c_smbus_write_word_swapped()
>>>     opt->ok_to_ignore_lock = true;
>>>             Now an interrupt comes
>>>                                      -> opt3001_irq
>>>                                        ..
>>>                                        opt->result_ready = true
>>>                                        wake_up()
>>>      opt->result_ready = false;
>>>      wait_event_timeout()
>>>
>>> In this case the first thread misses the result and waits until timeout expires.
>>>
>> Agreed - looks like some reordering is needed here as well.
>>
>> Jonathan
>>
> 
> In opt3001_get_lux has a comment, that i2c_smbus_write_word_swapped
> (line 246) enables interrupt mechanism. If an interrupt can not arise
> before the function, the assignments to both of flags should be moved
> before i2c_smbus_write_word_swapped and this is the best fix for both
> of issues. Do you know if my assumption is correct and interrupts are
> disabled before i2c_smbus_write_word_swapped call?

Andreas, can you confirm this for us?

Thanks,

Jonathan

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web