[meta-intel] Partprobe hang at boot for edison builds

[Andrei Gherzan]

Hello guys,

I'm facing a very strange bug on Edison board using the Intel Edison
Breakout board: sometimes, partprobe hangs at boot.

In order to reproduce the behavior I compiled a new vanilla Intel build
using source code and instructions from git://
git.yoctoproject.org/meta-intel-edison (0c8a23e linux: fix the
do_kernel_checkout error in src building). After this I added a service as
it follows to catch a hang:

$ cat /lib/systemd/system/test.service
[Unit]
Description=Test service
Requires=resin-init.service
After=resin-init.service
[Service]
ExecStart=/usr/sbin/partprobe ; /sbin/reboot
Type=oneshot
RemainAfterExit=yes
[Install]
WantedBy=basic.target

When partprobe hangs, the service shows the hang:
root at edison:~# systemctl status test
��● test.service - Test service
   Loaded: loaded (/lib/systemd/system/test.service; enabled)
   Active: activating (start) since Thu 2015-07-23 00:49:36 UTC; 19min ago
 Main PID: 186 (partprobe)
   CGroup: /system.slice/test.service
           ��└��─186 /usr/sbin/partprobe

As well, after a while on serial kernel dumps these messages:
[  240.632138] INFO: task mmcqd/0boot0:70 blocked for more than 120 seconds.
[  240.632233] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables
this message.
[  240.632982] INFO: task partprobe:186 blocked for more than 120 seconds.

Relevant or not, when the board doesn't hang, if I run partprobe manually
it throws errors as it follows:
root at edison:~# partprobe
[ 1196.432197] end_request: I/O error, dev mmcblk0rpmb, sector 0
[ 1196.496034] end_request: I/O error, dev mmcblk0rpmb, sector 0
Warning: Error fsyncing/closing /dev/mmcblk0rpmb: Input/output error

This is a blocking issue for us (at resin.io) because we suspect that the
underlying cause of this issue is affecting us in another bug that is way
harder to explain or to reproduce. Have you ever seen something like this?
I would gladly help in any way possible for debugging and fixing this bug.
Right now the only lead I had was to use a patch similar to:
http://permalink.gmane.org/gmane.linux.kernel.mmc/24260 . I booted a kernel
with this patch and nothing changes regarding to this bug. As well I'd like
to mention that these behaviors are reproducible on multiple boards - just
to rule out the hardware issue possibility.

Regards,
--
Andrei Gherzan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.yoctoproject.org/pipermail/meta-intel/attachments/20150723/44084bea/attachment-0001.html>