RT-Kernel boot stalls because of change in fs/dcache.c

RT-Kernel boot stalls because of change in fs/dcache.c

[Schrempf Frieder]

Hi,

I have tried to boot a 4.14-Kernel with the RT-patches and 
PREEMPT_RT_FULL on an i.MX6UL board. Unfortunately the boot stalls at 
some point and never finishes.

I did some bisecting and found out, that reverting this change: [1], 
makes the board boot correctly.

Can anyone help to come up with a proper fix for this?

Thanks,
Frieder

[1]: 
https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v4.14-rt&id=56684d21f7f8341f7a5ed1433714112f49294590

Re: RT-Kernel boot stalls because of change in fs/dcache.c

[Schrempf Frieder]

+ linux-rt-users

On 17.12.18 11:42, Frieder Schrempf wrote:
> Hi,
> 
> I have tried to boot a 4.14-Kernel with the RT-patches and 
> PREEMPT_RT_FULL on an i.MX6UL board. Unfortunately the boot stalls at 
> some point and never finishes.
> 
> I did some bisecting and found out, that reverting this change: [1], 
> makes the board boot correctly.
> 
> Can anyone help to come up with a proper fix for this?
> 
> Thanks,
> Frieder
> 
> [1]: 
> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v4.14-rt&id=56684d21f7f8341f7a5ed1433714112f49294590 
> 

Re: RT-Kernel boot stalls because of change in fs/dcache.c

[Sebastian Andrzej Siewior]

On 2018-12-19 07:50:17 [+0000], Schrempf Frieder wrote:
> + linux-rt-users
> 
> On 17.12.18 11:42, Frieder Schrempf wrote:
> > Hi,
> > 
> > I have tried to boot a 4.14-Kernel with the RT-patches and 
> > PREEMPT_RT_FULL on an i.MX6UL board. Unfortunately the boot stalls at 
> > some point and never finishes.
> > 
> > I did some bisecting and found out, that reverting this change: [1], 
> > makes the board boot correctly.
> > 
> > Can anyone help to come up with a proper fix for this?

can you check if the hashtable is initialized twice?

> > Thanks,
> > Frieder
> > 
> > [1]: 
> > https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v4.14-rt&id=56684d21f7f8341f7a5ed1433714112f49294590 

Sebastian

Re: RT-Kernel boot stalls because of change in fs/dcache.c

[Schrempf Frieder]

Hi Sebastian,

On 19.12.18 12:36, Sebastian Andrzej Siewior wrote:
> On 2018-12-19 07:50:17 [+0000], Schrempf Frieder wrote:
>> + linux-rt-users
>>
>> On 17.12.18 11:42, Frieder Schrempf wrote:
>>> Hi,
>>>
>>> I have tried to boot a 4.14-Kernel with the RT-patches and
>>> PREEMPT_RT_FULL on an i.MX6UL board. Unfortunately the boot stalls at
>>> some point and never finishes.
>>>
>>> I did some bisecting and found out, that reverting this change: [1],
>>> makes the board boot correctly.
>>>
>>> Can anyone help to come up with a proper fix for this?
> 
> can you check if the hashtable is initialized twice?

I'm not sure how to check, but I guess you are on the right track. I can 
leave the init loop for dentry_hashtable in place and remove the one for 
in_lookup_hashtable in vfs_caches_init_early() and this also makes the 
board boot.

Could it be, that the initializing for dentry_hashtable and 
in_lookup_hashtable somehow interfere?

Thanks for your help,
Frieder

> 
>>> Thanks,
>>> Frieder
>>>
>>> [1]:
>>> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v4.14-rt&id=56684d21f7f8341f7a5ed1433714112f49294590
> 
> Sebastian
> 

Re: RT-Kernel boot stalls because of change in fs/dcache.c

[Schrempf Frieder]

On 19.12.18 14:46, Frieder Schrempf wrote:
> Hi Sebastian,
> 
> On 19.12.18 12:36, Sebastian Andrzej Siewior wrote:
>> On 2018-12-19 07:50:17 [+0000], Schrempf Frieder wrote:
>>> + linux-rt-users
>>>
>>> On 17.12.18 11:42, Frieder Schrempf wrote:
>>>> Hi,
>>>>
>>>> I have tried to boot a 4.14-Kernel with the RT-patches and
>>>> PREEMPT_RT_FULL on an i.MX6UL board. Unfortunately the boot stalls at
>>>> some point and never finishes.
>>>>
>>>> I did some bisecting and found out, that reverting this change: [1],
>>>> makes the board boot correctly.
>>>>
>>>> Can anyone help to come up with a proper fix for this?
>>
>> can you check if the hashtable is initialized twice?
> 
> I'm not sure how to check, but I guess you are on the right track. I can 
> leave the init loop for dentry_hashtable in place and remove the one for 
> in_lookup_hashtable in vfs_caches_init_early() and this also makes the 
> board boot.
> 
> Could it be, that the initializing for dentry_hashtable and 
> in_lookup_hashtable somehow interfere?

So the source of the issue turned out to be completely unrelated to 
this. Although it seems like enabling PREEMPT_RT_FULL triggered the 
problem in my case.

The lockup of the kernel was caused by the SNVS RTC driver. Pulling in 
this patch: [1], solved the problem.

As this patch fixes a real bug I sent a request for it to be backported 
to the stable releases.

[1]: 
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=cd7f3a249dbed2858e6c2f30e5be7f1f7a709ee2

>>>>
>>>> [1]:
>>>> https://git.kernel.org/pub/scm/linux/kernel/git/rt/linux-stable-rt.git/commit/?h=v4.14-rt&id=56684d21f7f8341f7a5ed1433714112f49294590 
>>>>
>>
>> Sebastian
>>