Re: [Xenomai-help] [RTnet-users] Alignment problem

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 573 bytes --]

Hi everybody,

I have try to solve my problem by using a newer kernel (2.6.30) without
success, I have always the same issue with rtnet when I try rtping :

#./rtping 127.0.0.1
Xenomai: suspending kernel thread bf009310 ('rtnet-rtpc') at 0xc002d3c8
after exception #0

I tried other kernel like 2.6.33 but I have a known problem with the
Ethernet controller (dm9000 NETDEV WATCHDOG timeout).

I would like to solve this problem, in /proc/xenomai/faults, the exception
#0 is describe as a data or instruction access error, What does it means ?

Best regards,

WALLOIS Cyril

[-- Attachment #2: Type: text/html, Size: 629 bytes --]

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Cyril Wallois wrote:
> Hi everybody,
> 
> I have try to solve my problem by using a newer kernel (2.6.30) without
> success, I have always the same issue with rtnet when I try rtping :
> 
> #./rtping 127.0.0.1
> Xenomai: suspending kernel thread bf009310 ('rtnet-rtpc') at 0xc002d3c8
> after exception #0

Why simply not disassembling the kernel and looking at what is at
0xc002d3c8 and looking at the values of the registers? It is entirely
possible that the RTnet IP stack requires unaligned accesses (the
vanilla kernel does it, notably for handling of IP options), and as you
may know it, unaligned accesses may require software emulation on some
platforms. However, I am not sure you would get an exception #0 in that
case. Your post does not even tell us on what platform you get this
issue, though I suspect this is ARM.

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 2742 bytes --]

Hi Gilles,

Sorry, I don't mention that my problem has ever been evoqued on xenomai-help
one month ago :

2010/6/8 Cyril Wallois wrote:
> Hi Gilles,
>
> Thanks for your help
>
>     Still on ARM, right? Do you have the same issue if you use an I-pipe
>     patch for a more recent kernel versions (anything more recent than
>     2.6.28).
>
>
> Yes on ARM with PXA270 (toradex colibri), I try to use a newer kernel
> but I have a problem of file system and can't use NFS (ethernet driver
> as module for rtnet).
> On an other embedded system, with kernel 2.6.29, I haven't any problem.
>
>     Other than that, you do not give us enough information to help solving
>     your issue. What about the last issue you had? What was the value of
the
>     r3 register at the time of the alignment fault?
>
>
> I don't solve the last issue, I focus on try with newer version of
> kernel and xenomai.
> Probably my problem should be solve with a more recent kernel , I think
> I will try 2.6.29 (and try to solve fs problem).

I do not understand why choosing such an old kernel when starting a new
project. Why not 2.6.33? 2.6.29 is more than one year old.

As for the unaligned access, it looks like a problem in rtnet. So, I
doubt it could be solved by a more recent version of Xenomai. For the
record, I have been using Xenomai and rtnet for a long time on ARM, and
had no issues, so, there is definitely no bug fixed recently in Xenomai
to get rtnet working on ARM, they have been working together for a long
time. But Ok, it will avoid bugs that we already know.

--
                                            Gilles.

2010/7/15 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>

> Cyril Wallois wrote:
> > Hi everybody,
> >
> > I have try to solve my problem by using a newer kernel (2.6.30) without
> > success, I have always the same issue with rtnet when I try rtping :
> >
> > #./rtping 127.0.0.1
> > Xenomai: suspending kernel thread bf009310 ('rtnet-rtpc') at 0xc002d3c8
> > after exception #0
>
> Why simply not disassembling the kernel and looking at what is at
> 0xc002d3c8 and looking at the values of the registers? It is entirely
> possible that the RTnet IP stack requires unaligned accesses (the
> vanilla kernel does it, notably for handling of IP options), and as you
> may know it, unaligned accesses may require software emulation on some
> platforms. However, I am not sure you would get an exception #0 in that
> case. Your post does not even tell us on what platform you get this
> issue, though I suspect this is ARM.
>
> --
>                                             Gilles.
>
0xc002d3c8 correspond to do_page_fault in /proc/kallsyms

I will looking at the value of the register.

Best regards,

WALLOIS Cyril

[-- Attachment #2: Type: text/html, Size: 3530 bytes --]

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Cyril Wallois wrote:
> Hi Gilles,
> 
> Sorry, I don't mention that my problem has ever been evoqued on
> xenomai-help one month ago :

Ok. But if you had used the "reply" function of your MUA, the mail you
would have sent would have contained a "Reference" header pointing to
the mail you were replying, which would have allowed the MUA of the
people reading your mail, including Xenomai mailing list archives, to
properly sort this new mail as a follow-up of the mail you were replying to.

> 0xc002d3c8 correspond to do_page_fault in /proc/kallsyms

Ok. So, you enter do_page_fault from primary mode. Compile your kernel
with frame pointers, and try and insert a call to show_stack(NULL,NULL)
in the Xenomai code printing the messages which you see.

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Cyril Wallois wrote:
> Thanks for your reply,
> 
> 2010/7/15 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org
> <mailto:gilles.chanteperdrix@xenomai.org>>
> 
>     Cyril Wallois wrote:
>     > Hi Gilles,
>     >
>     > Sorry, I don't mention that my problem has ever been evoqued on
>     > xenomai-help one month ago :
> 
>     Ok. But if you had used the "reply" function of your MUA, the mail you
>     would have sent would have contained a "Reference" header pointing to
>     the mail you were replying, which would have allowed the MUA of the
>     people reading your mail, including Xenomai mailing list archives, to
>     properly sort this new mail as a follow-up of the mail you were
>     replying to.
> 
> The reply function send email to your address, so I thought that you
> should be the only one person who could read it.

I do not know about your MUA, but mine has a "Reply all" button, which
sets the references correctly as well, and allows to add new addresses
to the mail.

The point is, you can not expect everybody which reads your post, and
which would find your post in the archives to remember what platform you
use. So, either use the "reply" function of a MUA so that mine allows me
to find back the context, or repeat what is your setup every time you
post to the mailing list.

>     Ok. So, you enter do_page_fault from primary mode. Compile your kernel
>     with frame pointers, and try and insert a call to show_stack(NULL,NULL)
>     in the Xenomai code printing the messages which you see.
> 
> I try show_stack(NULL,NULL)  after the Xenomai error message, but I have
> a lock up (the Xenomai code is not printing).

Ok. Try adding a call to xnarch_trace_panic_freeze() before the call to
show_stack.

> What do you mean by primary mode, do you refer to real-time domain ?

Right, primary mode is a term specific to user-space threads.

P.S: The next private mail you send me will receive no answer.

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Gilles Chanteperdrix wrote:
> Cyril Wallois wrote:
>> I try show_stack(NULL,NULL)  after the Xenomai error message, but I have
>> a lock up (the Xenomai code is not printing).
> 
> Ok. Try adding a call to xnarch_trace_panic_freeze() before the call to
> show_stack.

Sorry, I meant rthal_emergency_console()

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 486 bytes --]

Thanks for all Gilles.

Cyril

2010/7/16 Gilles Chanteperdrix <gilles.chanteperdrix@xenomai.org>

> Gilles Chanteperdrix wrote:
> > Cyril Wallois wrote:
> >> I try show_stack(NULL,NULL)  after the Xenomai error message, but I have
> >> a lock up (the Xenomai code is not printing).
> >
> > Ok. Try adding a call to xnarch_trace_panic_freeze() before the call to
> > show_stack.
>
> Sorry, I meant rthal_emergency_console()
>
> --
>                                             Gilles.
>

[-- Attachment #2: Type: text/html, Size: 874 bytes --]

[Xenomai-help] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 905 bytes --]

Hello everybody,

I try rtnet with the last stable version of xenomai (2.5.3 and linux kernel
2.6.26) and I have a new error code :

./rtping 127.0.0.1
Real-time PING 127.0.0.1 56(84) bytes of data.
[  610.438400] Xenomai: suspending kernel thread bf027038 ('rtnet-rtpc') at
0xc0030fa4 after exception #0

/proc/xenomai/faults :
TRAP         CPU0
  0:            1    (Data or instruction access)
  1:            0    (Section fault)
  2:            0    (Generic data abort)
  3:            0    (Unknown exception)
  4:            0    (Instruction breakpoint)
  5:            0    (Floating point exception)
  6:            0    (VFP Floating point exception)
  7:            0    (Undefined instruction)
  8:            0    (Unaligned access exception)


in /proc/kallsyms 0xc0030fa4 correspond to : c0030f60 t do_page_fault

and bf027038 b dispatch_task        [rtnet]

Best regards,

WALLOIS Cyril

[-- Attachment #2: Type: text/html, Size: 1013 bytes --]

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Cyril Wallois wrote:
> Hello everybody,
> 
> I try rtnet with the last stable version of xenomai (2.5.3 and linux
> kernel 2.6.26) and I have a new error code :

Still on ARM, right? Do you have the same issue if you use an I-pipe
patch for a more recent kernel versions (anything more recent than 2.6.28).

Other than that, you do not give us enough information to help solving
your issue. What about the last issue you had? What was the value of the
r3 register at the time of the alignment fault?

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 852 bytes --]

Hi Gilles,

Thanks for your help

Still on ARM, right? Do you have the same issue if you use an I-pipe
> patch for a more recent kernel versions (anything more recent than 2.6.28).
>

Yes on ARM with PXA270 (toradex colibri), I try to use a newer kernel but I
have a problem of file system and can't use NFS (ethernet driver as module
for rtnet).
On an other embedded system, with kernel 2.6.29, I haven't any problem.

Other than that, you do not give us enough information to help solving
> your issue. What about the last issue you had? What was the value of the
> r3 register at the time of the alignment fault?
>

I don't solve the last issue, I focus on try with newer version of kernel
and xenomai.
Probably my problem should be solve with a more recent kernel , I think I
will try 2.6.29 (and try to solve fs problem).

Regards,

WALLOIS Cyril

[-- Attachment #2: Type: text/html, Size: 1318 bytes --]

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Gilles Chanteperdrix]

Cyril Wallois wrote:
> Hi Gilles,
> 
> Thanks for your help
> 
>     Still on ARM, right? Do you have the same issue if you use an I-pipe
>     patch for a more recent kernel versions (anything more recent than
>     2.6.28).
> 
>  
> Yes on ARM with PXA270 (toradex colibri), I try to use a newer kernel
> but I have a problem of file system and can't use NFS (ethernet driver
> as module for rtnet).
> On an other embedded system, with kernel 2.6.29, I haven't any problem.
> 
>     Other than that, you do not give us enough information to help solving
>     your issue. What about the last issue you had? What was the value of the
>     r3 register at the time of the alignment fault?
> 
>  
> I don't solve the last issue, I focus on try with newer version of
> kernel and xenomai.
> Probably my problem should be solve with a more recent kernel , I think
> I will try 2.6.29 (and try to solve fs problem).

I do not understand why choosing such an old kernel when starting a new
project. Why not 2.6.33? 2.6.29 is more than one year old.

As for the unaligned access, it looks like a problem in rtnet. So, I
doubt it could be solved by a more recent version of Xenomai. For the
record, I have been using Xenomai and rtnet for a long time on ARM, and
had no issues, so, there is definitely no bug fixed recently in Xenomai
to get rtnet working on ARM, they have been working together for a long
time. But Ok, it will avoid bugs that we already know.

-- 
					    Gilles.

Re: [Xenomai-help] [RTnet-users] Alignment problem

[Cyril Wallois]

[-- Attachment #1: Type: text/plain, Size: 204 bytes --]

Thanks for your answer Gilles,

I would try with 2.6.29 because I use this kernel on an other
Arm system (and it works).
I will try with the newer kernel supporting by this card.

Regards,

WALLOIS Cyril

[-- Attachment #2: Type: text/html, Size: 233 bytes --]