[ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[Oliver Munz]

Hi

To debug an USB-driver, I used diag_printf(). Unfortunatly this function 
blocks my DSR longer then the USB-specifications allowed to. So I tryed to 
send the debug-output over the standard at91-serial-driver, configurated 
with a 2kByte buffer.

This works fine a short time and fail:

CYG_ASSERT( ((queue_map & (1<<pri))!=0) == ((!run_queue[pri].empty())!=0), 
"Map and queue disagree");
in:

7 cyg_assert_fail() at e:\ecos\packages\infra\current\src\buffer.cxx:737
6 Cyg_Scheduler_Implementation::rem_thread() at 
e:\ecos\packages\kernel\current\src\sched\mlqueue.cxx:323
5 Cyg_Condition_Variable::wait_inner() at 
e:\ecos\packages\kernel\current\src\sync\mutex.cxx:636
4 serial_write() at 
e:\ecos\packages\io\serial\current\src\common\serial.c:362
3 cyg_io_write() at e:\ecos\packages\io\common\current\src\iosys.c:178


I don't know if it's allowed to use the buffered IO from ISR and DSR's. If 
everybody know's if it should work or not in threory, let me know.... Im 
also intressted in other hints how to debug driveres...

Thanks
Oliver Munz 


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[Fabian Scheler]

Hi,

> I don't know if it's allowed to use the buffered IO from ISR and DSR's. If
> everybody know's if it should work or not in threory, let me know.... Im
> also intressted in other hints how to debug driveres...

well, I don't know the exact implementation of buffered IO in eCos,
but buffered IO always is blocking (if you want to ensure that no data
gets lost), and blokcing operations are never allowed within an ISR or
a DSR.

Ciao, Fabian

--
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[Andrew Lunn]

On Tue, Jul 05, 2005 at 09:16:44AM +0200, Fabian Scheler wrote:
> Hi,
> 
> > I don't know if it's allowed to use the buffered IO from ISR and DSR's. If
> > everybody know's if it should work or not in threory, let me know.... Im
> > also intressted in other hints how to debug driveres...
> 
> well, I don't know the exact implementation of buffered IO in eCos,
> but buffered IO always is blocking (if you want to ensure that no data
> gets lost), and blokcing operations are never allowed within an ISR or
> a DSR.

Correct.

You can use the infrastructure trace functions and instramentation
functions which are safe when writting to a buffer. Or you can
implement your own buffering which is ISR/DSR safe.

        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[oliver munz @ s p e a g]

----- Original Message ----- 
From: "Andrew Lunn" <andrew@lunn.ch>
To: "Fabian Scheler" <fabian.scheler@gmail.com>
Cc: "Oliver Munz" <oli@snr.ch>; <ecos-discuss@sources.redhat.com>
Sent: Tuesday, July 05, 2005 9:27 AM
Subject: Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ...
Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()


> On Tue, Jul 05, 2005 at 09:16:44AM +0200, Fabian Scheler wrote:
>> Hi,
>>
>> > I don't know if it's allowed to use the buffered IO from ISR and DSR's.
>> > If
>> > everybody know's if it should work or not in threory, let me know....
>> > Im
>> > also intressted in other hints how to debug driveres...
>>
>> well, I don't know the exact implementation of buffered IO in eCos,
>> but buffered IO always is blocking (if you want to ensure that no data
>> gets lost), and blokcing operations are never allowed within an ISR or
>> a DSR.
>
> Correct.
>
> You can use the infrastructure trace functions and instramentation
> functions which are safe when writting to a buffer. Or you can
> implement your own buffering which is ISR/DSR safe.
>
>        Andrew

Thanks a lot...

Then I can try the option "CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING" (and hope
there rise no other problems)...

If I use the "CYG_TRACE0(bool, msg)" function family, then I have to use
"CYG_TRACE_PRINT()" if I want control over the timing, because
"CYG_TRACE_PRINT()" blocks until the hole trace-buffer ist written to
diag_write(). Right?

Greatings Oli


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[Andrew Lunn]

On Tue, Jul 05, 2005 at 11:43:20AM +0200, oliver munz @ s p e a g wrote:
> 
> ----- Original Message ----- 
> From: "Andrew Lunn" <andrew@lunn.ch>
> To: "Fabian Scheler" <fabian.scheler@gmail.com>
> Cc: "Oliver Munz" <oli@snr.ch>; <ecos-discuss@sources.redhat.com>
> Sent: Tuesday, July 05, 2005 9:27 AM
> Subject: Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ...
> Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()
> 
> 
> >On Tue, Jul 05, 2005 at 09:16:44AM +0200, Fabian Scheler wrote:
> >>Hi,
> >>
> >>> I don't know if it's allowed to use the buffered IO from ISR and DSR's.
> >>> If
> >>> everybody know's if it should work or not in threory, let me know....
> >>> Im
> >>> also intressted in other hints how to debug driveres...
> >>
> >>well, I don't know the exact implementation of buffered IO in eCos,
> >>but buffered IO always is blocking (if you want to ensure that no data
> >>gets lost), and blokcing operations are never allowed within an ISR or
> >>a DSR.
> >
> >Correct.
> >
> >You can use the infrastructure trace functions and instramentation
> >functions which are safe when writting to a buffer. Or you can
> >implement your own buffering which is ISR/DSR safe.
> >
> >       Andrew
> 
> Thanks a lot...
> 
> Then I can try the option "CYGOPT_IO_SERIAL_SUPPORT_NONBLOCKING" (and hope
> there rise no other problems)...

Nope, this will not help. Thats a different meaning of none blocking.

> If I use the "CYG_TRACE0(bool, msg)" function family, then I have to use
> "CYG_TRACE_PRINT()" if I want control over the timing, because
> "CYG_TRACE_PRINT()" blocks until the hole trace-buffer ist written to
> diag_write(). Right?

Correct. 

        Andrew

-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[Frank Pagliughi]

Oliver Munz wrote:

> Hi
>
> To debug an USB-driver, I used diag_printf(). Unfortunatly this 
> function blocks my DSR longer then the USB-specifications allowed to. 
> So I tryed to send the debug-output over the standard 
> at91-serial-driver, configurated with a 2kByte buffer.
>
> This works fine a short time and fail:
>
> CYG_ASSERT( ((queue_map & (1<<pri))!=0) == 
> ((!run_queue[pri].empty())!=0), "Map and queue disagree");
> in:
>
> 7 cyg_assert_fail() at e:\ecos\packages\infra\current\src\buffer.cxx:737
> 6 Cyg_Scheduler_Implementation::rem_thread() at 
> e:\ecos\packages\kernel\current\src\sched\mlqueue.cxx:323
> 5 Cyg_Condition_Variable::wait_inner() at 
> e:\ecos\packages\kernel\current\src\sync\mutex.cxx:636
> 4 serial_write() at 
> e:\ecos\packages\io\serial\current\src\common\serial.c:362
> 3 cyg_io_write() at e:\ecos\packages\io\common\current\src\iosys.c:178
>
>
> I don't know if it's allowed to use the buffered IO from ISR and 
> DSR's. If everybody know's if it should work or not in threory, let me 
> know.... Im also intressted in other hints how to debug driveres...
>
> Thanks
> Oliver Munz
>
Another thing you can consider is make a first pass at the driver using 
a high-priority thread to service the device, instead of the DSR. Just 
have the DSR signal the thread using a semaphore or similar. Than you 
can then send a fair amount of debug output to a hardware-driven serial 
port (from the thread context)  and still remain within the timing 
constraints of USB. You may just need to try responding to the USB host 
before doing your printf()'s in certain time-constrained situations, 
such as when you get a SET_ADDRESS.

This will obviously effect your throughput (not to mention all the 
serial output slowing things down) but you'll be able to get through a 
USB enumeration no problem. Once it's all working you can bring the code 
back into the DSR and remove the printing. Or even better, leave a 
configuration option to switch between the "debug" and "release" options.

Frank Pagliughi


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss

Re: [ECOS] Timecritical interrupt-debuging ... printf in DSR's ... Cyg_Scheduler_Implementation::rem_thread() ... cyg_assert_fail()

[oliver munz @ s p e a g]

Thanks for Your hints.

The soulution whit the high-priority thread is interessting. At the moment
I'm happy whit printf() on a standard buffered serial-line. When set to
non-blocking-mode via "retcode = cyg_io_set_config(comms[2],
CYG_IO_SET_CONFIG_SERIAL_WRITE_BLOCKING, &m, &n);" I lose characters, but
the DSR is running whitout timing- or kernel-problems...

Greatings
Oliver Munz


> Another thing you can consider is make a first pass at the driver using a
> high-priority thread to service the device, instead of the DSR. Just have
> the DSR signal the thread using a semaphore or similar. Than you can then
> send a fair amount of debug output to a hardware-driven serial port (from
> the thread context)  and still remain within the timing constraints of
> USB. You may just need to try responding to the USB host before doing your
> printf()'s in certain time-constrained situations, such as when you get a
> SET_ADDRESS.
>
> This will obviously effect your throughput (not to mention all the serial
> output slowing things down) but you'll be able to get through a USB
> enumeration no problem. Once it's all working you can bring the code back
> into the DSR and remove the printing. Or even better, leave a
> configuration option to switch between the "debug" and "release" options.
>
> Frank Pagliughi
>


-- 
Before posting, please read the FAQ: http://ecos.sourceware.org/fom/ecos
and search the list archive: http://ecos.sourceware.org/ml/ecos-discuss