fedora-xo 20090510 does not start X server

[Mikus Grinbergs]

Installed ~cjb/rawhide-xo 20090510.img with 'copy-nand' onto an XO-1 
system.  Booted.  Boot process went normally up to the point where X 
would be used to draw the background and the "user logon" screen.

Instead, the contents of the text console were repeatedly shown, 
blanked, shown, blanked, etc. - as the system repeatedly tried to 
start X, but failed.  Eventually, the boot process stalled at that 
point.

By switching to another text console, I was able to log in and 
examine the contents of memory.  The only thing of interest was many 
many copies (to all intents and purposed identical) of Xorg._.log in 
/var/log .  Copied the most recent of those, and put it into 
https://bugzilla.redhat.com/show_bug.cgi?id=500086 , which I created 
for that purpose.

mikus

[Peter Robinson]

Thanks for reporting that Mikus,

I'm seeing exactly the same problem on my XO. I've added some notes to
the bug as well as adding as a F11 blocker.

Peter

On Sun, May 10, 2009 at 8:31 PM, Mikus Grinbergs <mikus(a)bga.com&gt; wrote:
>  Installed ~cjb/rawhide-xo 20090510.img with 'copy-nand' onto
> an XO-1 system.
>   Booted.  Boot process went normally up to the point where X would be used
>  to draw the background and the "user logon" screen.
>
>  Instead, the contents of the text console were repeatedly shown, blanked,
>  shown, blanked, etc. - as the system repeatedly tried to start X, but
>  failed.  Eventually, the boot process stalled at that point.
>
>  By switching to another text console, I was able to log in and examine the
>  contents of memory.  The only thing of interest was many many copies (to all
>  intents and purposed identical) of Xorg._.log in /var/log .  Copied the most
>  recent of those, and put it into
>  https://bugzilla.redhat.com/show_bug.cgi?id=500086 , which I created for
>  that purpose.
>
>  mikus
>
>
>  _______________________________________________
>  Fedora-olpc-list mailing list
>  Fedora-olpc-list(a)redhat.com
>  https://www.redhat.com/mailman/listinfo/fedora-olpc-list
>

[Chris Ball]

Hi Mikus/Peter,

>  Installed ~cjb/rawhide-xo 20090510.img with 'copy-nand' onto
> an
>  XO-1 system.  Booted.  Boot process went normally up to the point
>  where X would be used to draw the background and the "user logon"
>  screen. 
I'm curious about how you managed to get that far -- on an XO here,
the 20090510.img prints "mount: unknown filesystem type 'jffs2'"
and doesn't progress further.  Does it not do the same for you?

Thanks,

- Chris.
-- 
Chris Ball   <cjb(a)laptop.org&gt;

[Chris Ball]

Hi,

>  I'm curious about how you managed to get that far -- on an XO
>  here, the 20090510.img prints "mount: unknown filesystem type
>  'jffs2'" and doesn't progress further.  Does it not do the same
>  for you? 
Oh, I see, it's some kind of race; a subsequent boot worked okay.

I think there are several bugs here:

* We get "mount: unknown filesystem type 'jffs2'" sporadically when
  booting, even though jffs2.ko is obviously in the initrd (since
  other boot attempts succeed).  I'm not sure which BZ component to
  file initrd bugs against.

* The already filed X bug.  (BZ #500086)

* When boot does progress, haldaemon sometimes fails to start, with a
  kernel protection fault in dmesg.  (Filed against kernel as BZ
  #500096.)

* Even when haldaemon does start, X doesn't seem to find any pointer
  devices.  Hard to find out more about this while X is segfaulting;
  I guess it's possible that the segfault is related..

- Chris.
-- 
Chris Ball   <cjb(a)laptop.org&gt;

[Peter Robinson]

Hi,

>    > I'm curious about how you managed to get that far -- on an
> XO
>    > here, the 20090510.img prints "mount: unknown filesystem type
>    > 'jffs2'" and doesn't progress further.  Does it not do the same
>    > for you?
>
>  Oh, I see, it's some kind of race; a subsequent boot worked okay.
>
>  I think there are several bugs here:
>
>  * We get "mount: unknown filesystem type 'jffs2'" sporadically when
>   booting, even though jffs2.ko is obviously in the initrd (since
>   other boot attempts succeed).  I'm not sure which BZ component to
>   file initrd bugs against. 
I would probably file one against either the kernel or mkinitrd. The
kernel guys are generally pretty quick to respond and reallocate it as
appropriate. The other thing to do is put the bug as a blocker to
"F11Blocker" as all of these are regressions from the last week or so
and IMO are definitely a blocker.

Peter

[Chris Ball]

Hi,

>  I would probably file one against either the kernel or
>  mkinitrd. 
Filed as https://bugzilla.redhat.com/show_bug.cgi?id=500196 , thanks.

- Chris.
-- 
Chris Ball   <cjb(a)laptop.org&gt;

[Peter Robinson]

>  Hi,
>
>    > I would probably file one against either the kernel or
>    > mkinitrd.
>
>  Filed as https://bugzilla.redhat.com/show_bug.cgi?id=500196 , thanks. 
Chris, out of interest how is the NAND attached wrt to bus etc to the
rest of the system. There's been some issues recently (mostly with usb
I think) with some of the fast boot features that have come from the
Intel Moblin patchset. I'm not sure whether this could be coming into
play or whether I have got completely the wrong end of the stick. This
link is quite a good article on LWN with further details.

http://lwn.net/Articles/330378/

Cheers,
Peter

[Chris Ball]

Hi,

>  Chris, out of interest how is the NAND attached wrt to bus etc to
>  the rest of the system. 
It speaks to our custom CaFE controller, via the "cafe_nand" kernel
module.

>  There's been some issues recently (mostly with usb I think) with
>  some of the fast boot features that have come from the Intel
>  Moblin patchset. I'm not sure whether this could be coming into
>  play or whether I have got completely the wrong end of the
>  stick. 
The part that has me most confused is that the specific error
message we get from mount should only ever appear if jffs2 is not in
/proc/filesystems, which should only ever happen if the kernel module
wasn't loaded.  The kernel module obviously is loaded sometimes, since
boot works sometimes, but somehow not on the failing boots.  I wonder
what's responsible for loading it; whether it's the initrd scripts or
mount itself that calls modprobe jffs2.

>  http://lwn.net/Articles/330378/ 
Thanks for the link.  We should try passing rootdelay= in olpc.fth,
even though the problem doesn't seem too similar.

- Chris.
-- 
Chris Ball   <cjb(a)laptop.org&gt;

[Peter Robinson]

>    > There's been some issues recently (mostly with usb I
> think) with
>    > some of the fast boot features that have come from the Intel
>    > Moblin patchset. I'm not sure whether this could be coming into
>    > play or whether I have got completely the wrong end of the
>    > stick.
>
>  The part that has me most confused is that the specific error
>  message we get from mount should only ever appear if jffs2 is not in
>  /proc/filesystems, which should only ever happen if the kernel module
>  wasn't loaded.  The kernel module obviously is loaded sometimes, since
>  boot works sometimes, but somehow not on the failing boots.  I wonder
>  what's responsible for loading it; whether it's the initrd scripts or
>  mount itself that calls modprobe jffs2. 
If you install the nash package (if its not there anyway) there's a
lsinitrd command that gives you the contents of the initrd package
including the initrd script that's run. Maybe there's been a change in
the order of the script or something.

Peter