Mounting large JFFS2 causes other process to stop running!

Mounting large JFFS2 causes other process to stop running!

[Benny Chen]

Hi there,
I have this problem, where when I mount a large JFFS2 partition (5M),
nothing else seems to be running.  The jffs2 routines seems to have the
processor all for itself.

I did a test by running 'top' and changed it's priority to -18 and then
mount the jffs2 partition.  Even with -18 priority 'top' stops reporting
for the duration of the mount.

I am aware that JFFS2 takes a long time to mount large partitions. But
to have no process running during the mount duration is a little funny.

Hope someone can help.

Regards,
Benny Chen

Re: Mounting large JFFS2 causes other process to stop running!

[David Jander]

On Monday 05 September 2005 04:42, Benny Chen wrote:
> Hi there,
> I have this problem, where when I mount a large JFFS2 partition (5M),
> nothing else seems to be running.  The jffs2 routines seems to have the
> processor all for itself.

What version of mtd/kernel are you using and on what hardware?

> I am aware that JFFS2 takes a long time to mount large partitions. But
> to have no process running during the mount duration is a little funny.

This sounds indeed funny.... almost tragic.
I have seen several minute-long stalls due to either mounting or GC, but all 
processes that are not dependant on that partition keep running nicely.
The only processes that suffer from such a thing are those that want to access 
the partition immediately, so mounting the root partition in rw-mode for 
example, is something to avoid because any given day GC can kick in right 
after mount, and then booting up can take veeery looong.

Greetings,

-- 
David Jander
Protonic Holland.

Re: Mounting large JFFS2 causes other process to stop running!

[David Jander]

On Monday 05 September 2005 10:55, Leon Zhang wrote:
> Hi. there.
> I confirm this bug.
> My kernel is linux-2.4.29-vrs1-pxa1-intc4-col1. and my board CPU is PXA27x,
> and the flash is on board flash.

I can't make up the MTD version out of your kernel version. Would it be the 
last CVS for 2.4.x from March 2005, or older?

> I am burned a minimum jffs2 rooffs. about 2M, only a busybox in it.
> But, if a burned a 10M size rootfs. The console respone slowly first, and
> after a while, the console will respone nonthing.

Can you turn debug-level to 1 in the jffs2 (and mtd) driver and post (relevant 
parts of) console output here? I'm not sure if I can help, but MTD developers 
might want to see what the kernel is busy with for so long.
I personally want to know if it has any relation with the architecture you are 
running on. I have had problems in the past with 2.4.x on PowerPC and older 
versions of MTD. Since I changed to CVS-march2005 I have not had any trouble, 
but I want to know more, and I need to be as sure as possible that I will not 
have any more trouble. The problem is, this kernel version is not officially 
supported anymore by mtd developers, so we are a little bit left out in the 
cold if there is trouble.

Best regards,

-- 
David Jander
Protonic Holland.

Re: Mounting large JFFS2 causes other process to stop running!

[Benny Chen]

Hi David,
Thanks for replying.

> I can't make up the MTD version out of your kernel version. Would it be the 
> last CVS for 2.4.x from March 2005, or older?

Yap, that is right. My mtd is quite old it is packaged with linux-2.4.21
kernel, so I checked the dates it is about 25/07/2002.  I guess
mtd-linux is not supporting 2.4.x anymore.  We will probably have to cop
this penalty until we upgrade our kernel.

Can you turn debug-level to 1 in the jffs2 (and mtd) driver and post
(relevant 
> parts of) console output here? I'm not sure if I can help, but MTD developers 
> might want to see what the kernel is busy with for so long.

Below are sniplets of the jffs2 debug level 1 message.  It loops through
the jffs2_scan_inode.node() for majority of the time.  This is also when
nothing else is running on the cpu.

Node is ino #2, version 43. Range 0x29000-0x2a000
jffs2_scan_inode_node(): Node at 0x0002ab80
Node is ino #2, version 44. Range 0x2a000-0x2b000
jffs2_scan_inode_node(): Node at 0x0002bbc4
Node is ino #2, version 45. Range 0x2b000-0x2c000
jffs2_scan_inode_node(): Node at 0x0002cc08
Node is ino #2, version 46. Range 0x2c000-0x2d000

Regards,
Benny

On Mon, 2005-09-05 at 21:46, David Jander wrote:
> On Monday 05 September 2005 10:55, Leon Zhang wrote:
> > Hi. there.
> > I confirm this bug.
> > My kernel is linux-2.4.29-vrs1-pxa1-intc4-col1. and my board CPU is PXA27x,
> > and the flash is on board flash.
> 
> I can't make up the MTD version out of your kernel version. Would it be the 
> last CVS for 2.4.x from March 2005, or older?
> 
> > I am burned a minimum jffs2 rooffs. about 2M, only a busybox in it.
> > But, if a burned a 10M size rootfs. The console respone slowly first, and
> > after a while, the console will respone nonthing.
> 
> Can you turn debug-level to 1 in the jffs2 (and mtd) driver and post (relevant 
> parts of) console output here? I'm not sure if I can help, but MTD developers 
> might want to see what the kernel is busy with for so long.
> I personally want to know if it has any relation with the architecture you are 
> running on. I have had problems in the past with 2.4.x on PowerPC and older 
> versions of MTD. Since I changed to CVS-march2005 I have not had any trouble, 
> but I want to know more, and I need to be as sure as possible that I will not 
> have any more trouble. The problem is, this kernel version is not officially 
> supported anymore by mtd developers, so we are a little bit left out in the 
> cold if there is trouble.
> 
> Best regards,

Re: Mounting large JFFS2 causes other process to stop running!

[David Woodhouse]

On Tue, 2005-09-06 at 15:52 +1000, Benny Chen wrote:
> Yap, that is right. My mtd is quite old it is packaged with linux-2.4.21
> kernel, so I checked the dates it is about 25/07/2002.  I guess
> mtd-linux is not supporting 2.4.x anymore.  We will probably have to cop
> this penalty until we upgrade our kernel.

If this is still a real problem in 2.4.31, then we would be interested
in fixing it. The background process _should_ be correctly yielding if
there are other processes to run.

-- 
dwmw2