rivatv / Bugs / #34 "Kernel level thread crashes"

Creature of Swamp - 2005-09-17

priority: 5 --> 7
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-09-17

Logged In: YES
user_id=1304731

The driver is showing the same behaviour as described
in the Bugreport "rivatv 8.3.0 on ELSA Erazor I PCI
freezes machine". I assume, that these bug reports are
related.
Additional information:
When calling v4l-info form the xawtv distribution,
v4l-info dumps all information to the commandline. A
second after the complete dump ( prompt is being
displayed ), the machine freezes. I assume, that there
must be a livelock, because you can hear the CPU fans
speed up to maximum speed. This happens, if the
CPUs are under load.
Additional environment information:
The kernel was compiled using gcc (GCC) 4.0.2
20050821 (prerelease).

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-19

assigned_to: nobody --> ela
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-19

Logged In: YES
user_id=80302

Hello!

Can you please try to change

I2C_CLASS_DDC into I2C_CLASS_HWMON

and/or

I2C_ADAP_CLASS_DDC into I2C_ADAP_CLASS_SMBUS

in the i2c_riva_bus0 structure in the i2c-riva.c file?

Probably this helps...

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-09-19

Logged In: YES
user_id=1304731

> Can you please try to change
> I2C_CLASS_DDC into I2C_CLASS_HWMON
> and/or
> I2C_ADAP_CLASS_DDC into
> I2C_ADAP_CLASS_SMBUS
> in the i2c_riva_bus0 structure in the i2c-riva.c file?

Sorry, none of the possible combinations helps. Driver
continues to crash. Perhaps hidden SMP problem????

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-20

Logged In: YES
user_id=80302

Yet another try: remove lock_kernel() and unlock_kernel() in
the 'i2c-riva.c' file...

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-09-20

Logged In: YES
user_id=1304731

> Yet another try: remove lock_kernel() and unlock_kernel() in
> the 'i2c-riva.c' file...

We seem to be on the right way ...
I removed the lock and unlock in the *original version* of
i2c-rova.c (No modifications to the constants of the first try.)

Now the modlues load cleanly without crashing.
BUT:
- v4l-info dumps all information and after a second the
computers hangs up.

- rmmod rivatv ; rmmod vpx32xx works in this order
BUT:
- rmmod vpx32xx ; rmmod rivatv causes my computer to hang
up after 2 seconds or something like that.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-21

Logged In: YES
user_id=80302

Hmpf. Probably also the spinlock_* functions? Try playing
with the LOCK_ macro... Probably you find a usable
combination...

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-09-21

Logged In: YES
user_id=1304731

It seems to me that is is getting difficulty.
There are two possibilities:
1. We are going to wait until 5th october. Then we are
going to have an online debug session. If you are going
to setup IPv6 you will be able to log into my machine. If
you need a tunnel broker search for "tunnel broker" in
google or use www.freenet6.net or www.sixxs.net . We
could use ICQ or talk in order to have interactive
discussion.
2. I am going to send you a card ( Elsa Victory Erazor
@ ebay 1 EUR + shipment ) as a donnation.

What do you want to do?

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-23

Logged In: YES
user_id=80302

Hello! Unfortunately I do not have access to the internet at
home and I can't misuse the access at work for this...
Probaly you may really try to remove

spin_lock_irqsave () from the LOCK_I2C_RIVA macro

and

spin_unlock_irqrestore () UNLOCK_I2C_RIVA macro

This and with the lock_kernel()/unlock_kernel() in place..

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-09-23

Logged In: YES
user_id=1304731

Hi!
Sorry, suggestion does not help ...
It seems to me that we have got stuck. This problem will
have to wait for two weeks. After that, I decided to
have an extended debugging session. Detailled plans:
1. Find out, what system calls are called form v4l-info.
2. Find out, which lock causes the machine to hang.
3. Find out, wheter we are dealing with an SMP problem
or not ( => test card on an non-SMP host )
4. Develop fix for this issuse ( perhaps )
At least I am going to provide more detailled debugging
information.
Is it possible to use the driver with kernel version 2.6.7?
The most comfortable way would be to use kgdb. kgdb
is only availible for kernel version 2.6.7.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Stefan Jahn - 2005-09-23

Logged In: YES
user_id=80302

Nice suggestion :-) Thanks for your efforts in advance!
RivaTV can be used with kernel 2.6.7... to answer your
question.

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

Creature of Swamp - 2005-10-22

Logged In: YES
user_id=1304731

Hi!
I had a debugging session last light and this morning. I was
analysing the code of v4l-info.
The results:
1. v4l-info opens /dev/video0, calls several times ioctl and
fillany crashes *after* the program has finished. After some
time of thinking, if found out, that the close system call
was missing in v4l-info. I.e the close system call causes
the system to hang.
Here is my test program (quick and very dirty):

#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <string.h>
#include <errno.h>
#include <fcntl.h>
#include <inttypes.h>
#include <ctype.h>
/*function used, in order to make sure, that there are no
side effects,
which could be mixed between different system calls.*/
void suspend() {
unsigned int count=90;
printf("Sleeping for 90 seconds ... \n");
for ( ; count > 0; count--){
sleep(1);
printf("%u\n", count);
}
printf("done. \n");
}

/*
---------------------------------------------------------------------
*/
/* main
*/
int main(int argc, char *argv[])
{
char dummy[256];
char *device = "/dev/video0";
int tab = 1, ok = 0;
int fd;

if (argc > 1)
device = argv[1];

printf("Running open system call ... ");
fd = open(device,O_RDONLY);

if (-1 == fd) {
fprintf(stderr,"open %s:
%s\n",device,strerror(errno));
exit(1);
};
printf("done.\n");
suspend();
printf("Running close system call ... ");
if(close(fd)){
fprintf(stderr,"Close system call failed.\n");
exit(1);
}
printf("done.\n");
suspend();

return 0;
}

The test Program prints:

Running open system call ... done
Sleeping for 90 seconds ...
90
89
88
...
0
done.
Running close system call ... done.
Sleeping for 90 seconds ...

Now my machine hangs . This means, that the close system
call is able to finish successfully, but something causes
the mahine to hang after the system call has finished.

Any ideas?

In order to a have a working kgdb setup, you will have to
wait for some time. There are still some problems which have
to be solved in my working environment.

Greetings
Frank

If you would like to refer to this comment somewhere else in this project, copy and paste the following link:

"Kernel level thread crashes"

Group

Searches

Help

#34 "Kernel level thread crashes"

Discussion