JFFS and RTLinux

JFFS and RTLinux

[Jan Debiec]

Hello,

I'm trying to use RealTimeLinux (RTL 3.0 pre9) on Arcom MediaGX SBC
(2.2.17 with MTD and JFFS1)
AS condition to run RTL, kernel (2.2.17) must be compiled with SMP flag
set on.
But then the JFFS is not running properly, kernel is not able to mount
the /dev/mtdblock2 as root fs. After message "Building JFFS fs crash
recovery tree ... " system is hanging.
Do you know the solution?

Jan
-- 
----------------------------------	
 Jan Debiec    
 Jurca Optoelektronik
 Raiffeisenstr. 5D
 D-63110 Rodgau (Dudenhofen)
 Germany 
 
 tel      +49 6106 8290 20
 fax      +49 6106 8290 26
 email    jan.debiec@jurca.de
 www.jurca.de
----------------------------------

RE: JFFS and RTLinux

[Vipin Malik]

I know that at least with RTAI, you can set the SMP flag off and then use
the 8259 timer UP scheduler. RTL does not have something similar?

Vipin

P.S. I assume that without the SMP flag (and RTLinux) everything works fine?

-----Original Message-----
From: Jan Debiec
To: linux-mtd@lists.infradead.org
Sent: 7/26/01 5:08 AM
Subject: JFFS and RTLinux

Hello,

I'm trying to use RealTimeLinux (RTL 3.0 pre9) on Arcom MediaGX SBC
(2.2.17 with MTD and JFFS1)
AS condition to run RTL, kernel (2.2.17) must be compiled with SMP flag
set on.
But then the JFFS is not running properly, kernel is not able to mount
the /dev/mtdblock2 as root fs. After message "Building JFFS fs crash
recovery tree ... " system is hanging.
Do you know the solution?

Jan
-- 
----------------------------------	
 Jan Debiec    
 Jurca Optoelektronik
 Raiffeisenstr. 5D
 D-63110 Rodgau (Dudenhofen)
 Germany 
 
 tel      +49 6106 8290 20
 fax      +49 6106 8290 26
 email    jan.debiec@jurca.de
 www.jurca.de
----------------------------------

______________________________________________________
Linux MTD discussion mailing list
http://lists.infradead.org/mailman/listinfo/linux-mtd/

Re: JFFS and RTLinux

[Jan Debiec]

Thanks Vipin,

I've installed RTAI-1.6 as UP with Arcom-2.2.17 kernel (by RTAI
configuration, you don't need to set SMP flag) and the Arcom board is
running. Not only board, the RTAI tests are running on Arcom too

Jan

P.S. But the question is still open, why has MTD and JFFS problems with
SMP flag set in kernel config.

Vipin Malik wrote:
> 
> I know that at least with RTAI, you can set the SMP flag off and then use
> the 8259 timer UP scheduler. RTL does not have something similar?
> 
> Vipin
> 
> P.S. I assume that without the SMP flag (and RTLinux) everything works fine?
> 
> -----Original Message-----
> From: Jan Debiec
> To: linux-mtd@lists.infradead.org
> Sent: 7/26/01 5:08 AM
> Subject: JFFS and RTLinux
> 
> Hello,
> 
> I'm trying to use RealTimeLinux (RTL 3.0 pre9) on Arcom MediaGX SBC
> (2.2.17 with MTD and JFFS1)
> AS condition to run RTL, kernel (2.2.17) must be compiled with SMP flag
> set on.
> But then the JFFS is not running properly, kernel is not able to mount
> the /dev/mtdblock2 as root fs. After message "Building JFFS fs crash
> recovery tree ... " system is hanging.
> Do you know the solution?
> 
> Jan
> --
> ----------------------------------
>  Jan Debiec
>  Jurca Optoelektronik
>  Raiffeisenstr. 5D
>  D-63110 Rodgau (Dudenhofen)
>  Germany
> 
>  tel      +49 6106 8290 20
>  fax      +49 6106 8290 26
>  email    jan.debiec@jurca.de
>  www.jurca.de
> ----------------------------------
> 
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/
> 
> ______________________________________________________
> Linux MTD discussion mailing list
> http://lists.infradead.org/mailman/listinfo/linux-mtd/

-- 
----------------------------------	
 Jan Debiec    
 Jurca Optoelektronik
 Raiffeisenstr. 5D
 D-63110 Rodgau (Dudenhofen)
 Germany 
 
 tel      +49 6106 8290 20
 fax      +49 6106 8290 26
 email    jan.debiec@jurca.de
 www.jurca.de
----------------------------------

Re: JFFS and RTLinux

[Vipin Malik]

Jan Debiec wrote:

> Thanks Vipin,
>
> I've installed RTAI-1.6 as UP with Arcom-2.2.17 kernel (by RTAI
> configuration, you don't need to set SMP flag) and the Arcom board is
> running. Not only board, the RTAI tests are running on Arcom too
>
> Jan
>
> P.S. But the question is still open, why has MTD and JFFS problems with
> SMP flag set in kernel config.

If I was forced to make a single guess, I would say that spin locks are probably
not being properly handled somewhere.
I cannot speak more intelligently beyond this :)

Vipin

Re: JFFS and RTLinux

[David Woodhouse]

jan.debiec@jurca.de said:
>  P.S. But the question is still open, why has MTD and JFFS problems
> with SMP flag set in kernel config. 

Reproduce on a non-RT kernel and use SysRq-P to find where it's blocking.

--
dwmw2

Re: JFFS and RTLinux

[Jan Debiec]

Hi David,

as you suggested,

David Woodhouse wrote:
> jan.debiec@jurca.de said:
> >  P.S. But the question is still open, why has MTD and JFFS problems
> > with SMP flag set in kernel config.
> 
> Reproduce on a non-RT kernel and use SysRq-P to find where it's blocking.


the results:
EIP: 0010:[<c01986d3>] EFLAGS: 00000202
...			EBP: c01d0040


in system.map for that kernel:
c0195940 T stext_lock
c019b360 R lk_lockmsg
...
c01d0040 d sbc_mediagx_map

unfortunately "stext_lock" tells me nothing (is used in head.S),
sbc_mediagx_map is a structure used for Arcom GXM board

Jan 
-- 
----------------------------------	
 Jan Debiec    
 Jurca Optoelektronik
 Raiffeisenstr. 5D
 D-63110 Rodgau (Dudenhofen)
 Germany 
 
 tel      +49 6106 8290 20
 fax      +49 6106 8290 26
 email    jan.debiec@jurca.de
 www.jurca.de
----------------------------------

Re: JFFS and RTLinux

[David Woodhouse]

jan.debiec@jurca.de said:
>  unfortunately "stext_lock" tells me nothing (is used in head.S),
> sbc_mediagx_map is a structure used for Arcom GXM board

Spinlock deadlock. Check the spin_lock() and spin_unlock() pairs are evenly 
matched in sbc_mediagx.c (they look like they are in the current version of 
sbc_gxx.c).

$ gdb vmlinux
(gdb) disassemble 0xc01986d3

It should be in a loop which looks something like

0xc0263620 <stext_lock+64>:	cmpb   $0x0,0xc02ccdf8
0xc0263627 <stext_lock+71>:	repz nop 
0xc0263629 <stext_lock+73>:	jle    0xc0263620 <stext_lock+64>
0xc026362b <stext_lock+75>:	jmp    0xc0105b3f <__down+111>

The place it's jumping to in the final line is the place that's trying to 
obtain the (blocked) spinlock. 

--
dwmw2