[Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Axel Beierlein]

what could it be that an simple rt_dev_open locks up my system when it  
calls rtdm_lock_put_irqrestore?

[...]
if (rtdm_irq_request(&ctx->irq_handle,
		        MPC5xxx_PSC2_IRQ,
			rt_psc_interrupt,
			RTDM_IRQTYPE_SHARED | RTDM_IRQTYPE_EDGE,
				context->device->proc_name,
				ctx))
	{
		printk ("psc5200B.c: request_irq() failed\n");
		return -1;
	}

     rtdm_irq_enable(&ctx->irq_handle);

     rtdm_lock_get_irqsave(&ctx->lock, lock_ctx);

     ctx->imr_status = MPC5xxx_PSC_IMR_RXRDY;
	out_be16(&psc->mpc5xxx_psc_imr, ctx->imr_status);

	rtdm_lock_put_irqrestore(&ctx->lock, lock_ctx);
	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
	Here i get the crash


The test prog is a simple  
fd=rt_dev_open("nameofdevice",0)......rt_dev_close(fd).


Axel

Re: [Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 1038 bytes --]

Axel Beierlein wrote:
> what could it be that an simple rt_dev_open locks up my system when it  
> calls rtdm_lock_put_irqrestore?
> 
> [...]
> if (rtdm_irq_request(&ctx->irq_handle,
> 		        MPC5xxx_PSC2_IRQ,
> 			rt_psc_interrupt,
> 			RTDM_IRQTYPE_SHARED | RTDM_IRQTYPE_EDGE,
> 				context->device->proc_name,
> 				ctx))
> 	{
> 		printk ("psc5200B.c: request_irq() failed\n");
> 		return -1;
> 	}
> 
>      rtdm_irq_enable(&ctx->irq_handle);
> 
>      rtdm_lock_get_irqsave(&ctx->lock, lock_ctx);
> 
>      ctx->imr_status = MPC5xxx_PSC_IMR_RXRDY;
> 	out_be16(&psc->mpc5xxx_psc_imr, ctx->imr_status);

Let me guess: this line arms the related IRQ source at hw level, no?

> 
> 	rtdm_lock_put_irqrestore(&ctx->lock, lock_ctx);
> 	^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 	Here i get the crash

Because it enabled local IRQ delivery again, and the system ends up in
an IRQ storm, or some endless loop in the IRQ handler, or some crash
related to IRQ handling. Still just guessing...

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2093 bytes --]

Axel Beierlein wrote:
> Am 29.08.2007, 16:42 Uhr, schrieb Jan Kiszka <jan.kiszka@domain.hid>:
> 
>> Axel Beierlein wrote:
>>> what could it be that an simple rt_dev_open locks up my system when it
>>> calls rtdm_lock_put_irqrestore?
>>>
>>> [...]
>>> if (rtdm_irq_request(&ctx->irq_handle,
>>>                 MPC5xxx_PSC2_IRQ,
>>>             rt_psc_interrupt,
>>>             RTDM_IRQTYPE_SHARED | RTDM_IRQTYPE_EDGE,
>>>                 context->device->proc_name,
>>>                 ctx))
>>>     {
>>>         printk ("psc5200B.c: request_irq() failed\n");
>>>         return -1;
>>>     }
>>>
>>>      rtdm_irq_enable(&ctx->irq_handle);
>>>
>>>      rtdm_lock_get_irqsave(&ctx->lock, lock_ctx);
>>>
>>>      ctx->imr_status = MPC5xxx_PSC_IMR_RXRDY;
>>>     out_be16(&psc->mpc5xxx_psc_imr, ctx->imr_status);
>>
>> Let me guess: this line arms the related IRQ source at hw level, no?
>>
> 
> Yes you are right. This Bit enables the RX_Ready IRQ at hw level. Sorry
> for the little Code fragment but the code relates to your 16550A.c and i
> don´t want to post all the routines at the moment.
> 
>>>
>>>     rtdm_lock_put_irqrestore(&ctx->lock, lock_ctx);
>>>     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>     Here i get the crash
>>
>> Because it enabled local IRQ delivery again, and the system ends up in
>> an IRQ storm, or some endless loop in the IRQ handler, or some crash
>> related to IRQ handling. Still just guessing...
>>
>> Jan
>>
> 
> But how your UART driver handle this Problem? I have check the Irq
> Handler an implement an printk in it to see where it hangs, but it
> doesn´t step into this handler.

In case you have an IRQ storm or your handler runs in an endless loop,
you won't be able to read that printk even if it's executed - Linux will
have no chance to dump the message.

Start debugging with shrinking the handler down to a plain "return
RTDM_IRQ_HANDLED" (or even RTDM_IRQ_NONE so that Xenomai disables the
line), then slowly re-enabling and reviewing your code.

Jan

PS: reply-to-all


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]

Re: [Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Axel Beierlein]

Am 30.08.2007, 08:55 Uhr, schrieb Jan Kiszka <jan.kiszka@domain.hid>:

> Axel Beierlein wrote:
>> Am 29.08.2007, 16:42 Uhr, schrieb Jan Kiszka <jan.kiszka@domain.hid>:
>>
>>> Axel Beierlein wrote:
>>>> what could it be that an simple rt_dev_open locks up my system when it
>>>> calls rtdm_lock_put_irqrestore?
>>>>
>>>> [...]
>>>> if (rtdm_irq_request(&ctx->irq_handle,
>>>>                 MPC5xxx_PSC2_IRQ,
>>>>             rt_psc_interrupt,
>>>>             RTDM_IRQTYPE_SHARED | RTDM_IRQTYPE_EDGE,
>>>>                 context->device->proc_name,
>>>>                 ctx))
>>>>     {
>>>>         printk ("psc5200B.c: request_irq() failed\n");
>>>>         return -1;
>>>>     }
>>>>
>>>>      rtdm_irq_enable(&ctx->irq_handle);
>>>>
>>>>      rtdm_lock_get_irqsave(&ctx->lock, lock_ctx);
>>>>
>>>>      ctx->imr_status = MPC5xxx_PSC_IMR_RXRDY;
>>>>     out_be16(&psc->mpc5xxx_psc_imr, ctx->imr_status);
>>>
>>> Let me guess: this line arms the related IRQ source at hw level, no?
>>>
>>
>> Yes you are right. This Bit enables the RX_Ready IRQ at hw level. Sorry
>> for the little Code fragment but the code relates to your 16550A.c and i
>> don´t want to post all the routines at the moment.
>>
>>>>
>>>>     rtdm_lock_put_irqrestore(&ctx->lock, lock_ctx);
>>>>     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>>     Here i get the crash
>>>
>>> Because it enabled local IRQ delivery again, and the system ends up in
>>> an IRQ storm, or some endless loop in the IRQ handler, or some crash
>>> related to IRQ handling. Still just guessing...
>>>
>>> Jan
>>>
>>
>> But how your UART driver handle this Problem? I have check the Irq
>> Handler an implement an printk in it to see where it hangs, but it
>> doesn´t step into this handler.
>
> In case you have an IRQ storm or your handler runs in an endless loop,
> you won't be able to read that printk even if it's executed - Linux will
> have no chance to dump the message.
>
> Start debugging with shrinking the handler down to a plain "return
> RTDM_IRQ_HANDLED" (or even RTDM_IRQ_NONE so that Xenomai disables the
> line), then slowly re-enabling and reviewing your code.

That has no effect. It doesn´t step into it.
I think there is another Problem, cause when i go back to the userspace  
applikation after i have opened the device the system locks up when i call  
a funktion like getchar()
Just in following dev_open and dev_close is workin for only one time. But  
when i call the same applikation a second time it locks up in the rt_open  
routine at rtdm_lock_put_irqrestore.

>
> Jan
>
> PS: reply-to-all

Ups, wrong button. Now i know why my message doesn´t appear in the mailing  
list ;-)


Axel

Re: [Xenomai-help] System hangs in/after rtdm_lock_put_irqrestore

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2931 bytes --]

Axel Beierlein wrote:
> Am 30.08.2007, 08:55 Uhr, schrieb Jan Kiszka <jan.kiszka@domain.hid>:
> 
>> Axel Beierlein wrote:
>>> Am 29.08.2007, 16:42 Uhr, schrieb Jan Kiszka <jan.kiszka@domain.hid>:
>>>
>>>> Axel Beierlein wrote:
>>>>> what could it be that an simple rt_dev_open locks up my system when it
>>>>> calls rtdm_lock_put_irqrestore?
>>>>>
>>>>> [...]
>>>>> if (rtdm_irq_request(&ctx->irq_handle,
>>>>>                 MPC5xxx_PSC2_IRQ,
>>>>>             rt_psc_interrupt,
>>>>>             RTDM_IRQTYPE_SHARED | RTDM_IRQTYPE_EDGE,
>>>>>                 context->device->proc_name,
>>>>>                 ctx))
>>>>>     {
>>>>>         printk ("psc5200B.c: request_irq() failed\n");
>>>>>         return -1;
>>>>>     }
>>>>>
>>>>>      rtdm_irq_enable(&ctx->irq_handle);
>>>>>
>>>>>      rtdm_lock_get_irqsave(&ctx->lock, lock_ctx);
>>>>>
>>>>>      ctx->imr_status = MPC5xxx_PSC_IMR_RXRDY;
>>>>>     out_be16(&psc->mpc5xxx_psc_imr, ctx->imr_status);
>>>>
>>>> Let me guess: this line arms the related IRQ source at hw level, no?
>>>>
>>>
>>> Yes you are right. This Bit enables the RX_Ready IRQ at hw level. Sorry
>>> for the little Code fragment but the code relates to your 16550A.c and i
>>> don´t want to post all the routines at the moment.
>>>
>>>>>
>>>>>     rtdm_lock_put_irqrestore(&ctx->lock, lock_ctx);
>>>>>     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
>>>>>     Here i get the crash
>>>>
>>>> Because it enabled local IRQ delivery again, and the system ends up in
>>>> an IRQ storm, or some endless loop in the IRQ handler, or some crash
>>>> related to IRQ handling. Still just guessing...
>>>>
>>>> Jan
>>>>
>>>
>>> But how your UART driver handle this Problem? I have check the Irq
>>> Handler an implement an printk in it to see where it hangs, but it
>>> doesn´t step into this handler.
>>
>> In case you have an IRQ storm or your handler runs in an endless loop,
>> you won't be able to read that printk even if it's executed - Linux will
>> have no chance to dump the message.
>>
>> Start debugging with shrinking the handler down to a plain "return
>> RTDM_IRQ_HANDLED" (or even RTDM_IRQ_NONE so that Xenomai disables the
>> line), then slowly re-enabling and reviewing your code.
> 
> That has no effect. It doesn´t step into it.
> I think there is another Problem, cause when i go back to the userspace
> applikation after i have opened the device the system locks up when i
> call a funktion like getchar()
> Just in following dev_open and dev_close is workin for only one time.
> But when i call the same applikation a second time it locks up in the
> rt_open routine at rtdm_lock_put_irqrestore.

Something must be very wrong. Do you register the correct IRQ line /
modify the right hw registers? Also, don't you have some hardware
debugger at hand to find out where the system goes to, what IRQs show up?

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]