Question about C language.

Question about C language.

[Francis Moreau]

Hello,

I know it's a bit out of topic but this is something I need to clarify for
writing a Linux driver... hope you don't mind.

In my driver I have a global variable that controls a loop such as:

int my_condition;

void change_my_condition(int new)
{
    my_condition = new;
}

int foo(void)
{
    /* irqs are disabled */
    my_condition = 1;
    do {
        ....
        local_irq_enable();
        cpu_sleep();
        local_irq_disable();

   } while (my_condition);

}

This variable is modified by an interrupt handler define in another file
by using 'change_my_condition' function.

By reading the ISO C99 specification, I _think_ that I needn't any
kind of barrier
or even use the volatile type qualifier for my_condition variable to make a true
access to 'my_condition' in the controlling expression of the while, but I'm not
sure.

Coud anybody confirm ?

Thanks,
-- 
Francis

Re: Question about C language.

[Chris Snook]

Francis Moreau wrote:
> Hello,
> 
> I know it's a bit out of topic but this is something I need to clarify for
> writing a Linux driver... hope you don't mind.
> 
> In my driver I have a global variable that controls a loop such as:
> 
> int my_condition;
> 
> void change_my_condition(int new)
> {
>     my_condition = new;
> }
> 
> int foo(void)
> {
>     /* irqs are disabled */
>     my_condition = 1;
>     do {
>         ....
>         local_irq_enable();
>         cpu_sleep();
>         local_irq_disable();
> 
>    } while (my_condition);
> 
> }
> 
> This variable is modified by an interrupt handler define in another file
> by using 'change_my_condition' function.
> 
> By reading the ISO C99 specification, I _think_ that I needn't any
> kind of barrier
> or even use the volatile type qualifier for my_condition variable to make a true
> access to 'my_condition' in the controlling expression of the while, but I'm not
> sure.
> 
> Coud anybody confirm ?
> 
> Thanks,

Even volatile may be insufficient with some architecture/compiler 
combinations.  You should use explicit barriers wherever you need them, 
or Bad Things will happen.

-- Chris

Re: Question about C language.

[Chris Snook]

Chris Snook wrote:
> Francis Moreau wrote:
>> Hello,
>>
>> I know it's a bit out of topic but this is something I need to clarify 
>> for
>> writing a Linux driver... hope you don't mind.
>>
>> In my driver I have a global variable that controls a loop such as:
>>
>> int my_condition;
>>
>> void change_my_condition(int new)
>> {
>>     my_condition = new;
>> }
>>
>> int foo(void)
>> {
>>     /* irqs are disabled */
>>     my_condition = 1;
>>     do {
>>         ....
>>         local_irq_enable();
>>         cpu_sleep();
>>         local_irq_disable();
>>
>>    } while (my_condition);
>>
>> }
>>
>> This variable is modified by an interrupt handler define in another file
>> by using 'change_my_condition' function.
>>
>> By reading the ISO C99 specification, I _think_ that I needn't any
>> kind of barrier
>> or even use the volatile type qualifier for my_condition variable to 
>> make a true
>> access to 'my_condition' in the controlling expression of the while, 
>> but I'm not
>> sure.
>>
>> Coud anybody confirm ?
>>
>> Thanks,
> 
> Even volatile may be insufficient with some architecture/compiler 
> combinations.  You should use explicit barriers wherever you need them, 
> or Bad Things will happen.
> 
> -- Chris
> 

Oops, forgot to mention, you should use atomic_t, to avoid aliasing 
problems, and ALSO use explicit barriers wherever you need them.

-- Chris

Re: Question about C language.

[Francis Moreau]

Hello,

On Sun, Mar 23, 2008 at 8:30 AM, Chris Snook <csnook@redhat.com> wrote:
>  > I know it's a bit out of topic but this is something I need to clarify for
>  > writing a Linux driver... hope you don't mind.
>  >
>  > In my driver I have a global variable that controls a loop such as:
>  >
>  > int my_condition;
>  >
>  > void change_my_condition(int new)
>  > {
>  >     my_condition = new;
>  > }
>  >
>  > int foo(void)
>  > {
>  >     /* irqs are disabled */
>  >     my_condition = 1;
>  >     do {
>  >         ....
>  >         local_irq_enable();
>  >         cpu_sleep();
>  >         local_irq_disable();
>  >
>  >    } while (my_condition);
>  >
>  > }
>  >
>  > This variable is modified by an interrupt handler define in another file
>  > by using 'change_my_condition' function.
>  >
>  > By reading the ISO C99 specification, I _think_ that I needn't any
>  > kind of barrier
>  > or even use the volatile type qualifier for my_condition variable to make a true
>  > access to 'my_condition' in the controlling expression of the while, but I'm not
>  > sure.
>  >
>  > Coud anybody confirm ?
>  >
>  > Thanks,
>
>  Even volatile may be insufficient with some architecture/compiler
>  combinations.  You should use explicit barriers wherever you need them,
>  or Bad Things will happen.
>

Really ?

Could you give me some details please ?

As I said previously, I don't think I need any barriers here because this code
doesn't have any reordering issues and futhermore only one CPU will be up
at this time.

Actually the only issue I can see is about compiler aliases. I don't know if
in that case it could use an alias for 'my_condition'. Looking at the code, it
would be a bad idea but I can't find anything about this in the C specification.
The C spec talks about sequence points but I'm really confused when reading
that part...

Thanks
-- 
Francis

Re: Question about C language.

[Francis Moreau]

On Sun, Mar 23, 2008 at 8:32 AM, Chris Snook <csnook@redhat.com> wrote:
>  Oops, forgot to mention, you should use atomic_t, to avoid aliasing
>  problems, and ALSO use explicit barriers wherever you need them.
>

Interesting, could you give me some pointers about aliasing please ?

Thanks
-- 
Francis