Re: [PATCH 4/4 v4] rtc: add rtc-driver for HID sensors of type time

[Lars-Peter Clausen]

On 12/14/2012 03:29 PM, Alexander Holler wrote:
> Am 14.12.2012 15:15, schrieb Alexander Holler:
>> Am 14.12.2012 14:08, schrieb Alexander Holler:
>>> Am 14.12.2012 10:42, schrieb Lars-Peter Clausen:
>>
>>>> And another thing I've overlooked before:
>>>> wait_for_completion_interruptible_timeout can either return a positive
>>>> number when the completion was completed, 0 in case of an timeout, or a
>>>> negative error code in case it was interrupted. You need to handle all
>>>> three. E.g. something like this.
>>>>
>>>> ret = wait_for_completion_interruptible_timeout(...)
>>>> if (ret == 0)
>>>>     return -EIO;
>>>> if (ret < 0)
>>>>     return ret
>>>>
>>>
>>> Hmpf, the only working approach to use some in kernel functions really
>>> is to the read source yourself and don't trust anything else. :/
>>
>> Anyway, my approach doesn't work as it introduces a race condition:
>>
>>
>> /* get a report with all values through requesting one value */
>> sensor_hub_input_attr_get_raw_value(...)
>>
>> /* race if this task goes to slepp and the values were
>> received before it could call the below wait...
>>
>> /* wait for all values (event) */
>> if (!wait_for_completion_interruptible_timeout(...))
>>
>>
>> I'll have to look for a mechanism how to avoid that. So v5 might need
>> some time.
> 
> Sorry for the noise. That INIT_COMPLETION() before the sensor...() does
> exactly that. So it's enough if I handle the different return situations of
> wait_for...().
> 
> I will just use if(wait...()<=0) return -EIO.
> 

No, that's wrong. You should really return the error code returned by
wait_for_completion_interruptible_timeout(). This will make sure that
userspace restarts the syscall if necessary.

- Lars
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/