[Xenomai] Make a block of code atomic

[Xenomai] Make a block of code atomic

[Paolo Minazzi]

Hi,

I would like to make a block of kernel code atomic, so that no one (both 
xenomai and Linux) can interrupt it.
I use the following :

#define my_local_irq_save(x)    ((x) = 
ipipe_test_and_stall_pipeline_head() & 1)
#define my_local_irq_restore(x) ipipe_restore_pipeline_head(x)

my_local_irq_save(flags);
< critical section >
my_local_irq_restore(flags);

It works well enough, but rarely an rtdm irq enter into my critical 
section.
The rtdm irq is registered in this way : rtdm_irq_request(&priv->irqHnd, 
ETH_PORT_IRQ_NUM(priv->port), irqHandler, RTDM_IRQTYPE_EDGE, "rtdm_eth", 
priv);

Is it normal ? Are you sure that it cannot happen ?

Thanks,
Paolo

Re: [Xenomai] Make a block of code atomic

[Gilles Chanteperdrix]

On 07/09/2013 01:40 PM, Paolo Minazzi wrote:

> Hi,
>
> I would like to make a block of kernel code atomic, so that no one (both
> xenomai and Linux) can interrupt it.
> I use the following :
>
> #define my_local_irq_save(x)    ((x) =
> ipipe_test_and_stall_pipeline_head()&  1)
> #define my_local_irq_restore(x) ipipe_restore_pipeline_head(x)
>
> my_local_irq_save(flags);
> <  critical section>
> my_local_irq_restore(flags);
>
> It works well enough, but rarely an rtdm irq enter into my critical
> section.
> The rtdm irq is registered in this way : rtdm_irq_request(&priv->irqHnd,
> ETH_PORT_IRQ_NUM(priv->port), irqHandler, RTDM_IRQTYPE_EDGE, "rtdm_eth",
> priv);
>
> Is it normal ? Are you sure that it cannot happen ?


Looking at ipipe_test_and_stall_head implementation on the I-pipe patch 
for Linux 3.8:

static inline unsigned long ipipe_test_and_stall_head(void)
{
	hard_local_irq_disable();
	return __test_and_set_bit(IPIPE_STALL_FLAG, &__ipipe_head_status);
}

getting an interrupt in your critical section would mean that 
"hard_local_irq_disable" is broken on the architecture where you use it.

That being said, I do not know if I look at the implementation you are 
using, since you did not give us any information about version and 
architecture you are using.

-- 
                                                                 Gilles.

Re: [Xenomai] Make a block of code atomic

[Paolo Minazzi]

Il 10/07/2013 13.37, Gilles Chanteperdrix ha scritto:
> On 07/09/2013 01:40 PM, Paolo Minazzi wrote:
>
>> Hi,
>>
>> I would like to make a block of kernel code atomic, so that no one (both
>> xenomai and Linux) can interrupt it.
>> I use the following :
>>
>> #define my_local_irq_save(x)    ((x) =
>> ipipe_test_and_stall_pipeline_head()&  1)
>> #define my_local_irq_restore(x) ipipe_restore_pipeline_head(x)
>>
>> my_local_irq_save(flags);
>> <  critical section>
>> my_local_irq_restore(flags);
>>
>> It works well enough, but rarely an rtdm irq enter into my critical
>> section.
>> The rtdm irq is registered in this way : rtdm_irq_request(&priv->irqHnd,
>> ETH_PORT_IRQ_NUM(priv->port), irqHandler, RTDM_IRQTYPE_EDGE, "rtdm_eth",
>> priv);
>>
>> Is it normal ? Are you sure that it cannot happen ?
>
>
> Looking at ipipe_test_and_stall_head implementation on the I-pipe 
> patch for Linux 3.8:
>
> static inline unsigned long ipipe_test_and_stall_head(void)
> {
>     hard_local_irq_disable();
>     return __test_and_set_bit(IPIPE_STALL_FLAG, &__ipipe_head_status);
> }
>
> getting an interrupt in your critical section would mean that 
> "hard_local_irq_disable" is broken on the architecture where you use it.
>
> That being said, I do not know if I look at the implementation you are 
> using, since you did not give us any information about version and 
> architecture you are using.
>
Hi Gilles, thanks for your answer.
I found this morning that was linux to re-enable interrupts.

The following

#define my_local_irq_save(x)    ((x) = 
ipipe_test_and_stall_pipeline_head()&  1)
#define my_local_irq_restore(x) ipipe_restore_pipeline_head(x)

works well if no one touch irq flags directly. Unfortunately was not my 
case.

Thanks again,
Paolo