Re: Unexpected switches to in-band

[Philippe Gerum <rpm@xenomai.org>]

Giulio Moro <giulio@bela.io> writes:

> I realised that gitlab's v6.12.y-evl-rebase and v6.1.y-cip-evl-rebase can boot as-is on the PocketBeagle 2 and so I'll keep using those for tests going forward, without any third-party patches.
>
>>> Ok, so it looks like both of you observe the same issue, which seems to
>>> be arch-independent. Checking the code which takes care of preventing
>>> COW for oob-enabled processes (a behavior which may cause unwanted minor
>>> faults in a tricky way), I stumbled upon a really bad bug. Could you
>>> check whether the patch below helps?
>>>
>>> diff --git a/include/linux/sched/coredump.h b/include/linux/sched/coredump.h
>>> index 73de18353e79..c6b1efcbd833 100644
>>> --- a/include/linux/sched/coredump.h
>>> +++ b/include/linux/sched/coredump.h
>>> @@ -91,7 +91,7 @@ static inline int get_dumpable(struct mm_struct *mm)
>>>     #define MMF_VM_MERGE_ANY	30
>>>   #define MMF_VM_MERGE_ANY_MASK	(1 << MMF_VM_MERGE_ANY)
>>> -#define MMF_DOVETAILED		31	/* mm belongs to a dovetailed process */
>>> +#define MMF_DOVETAILED		18	/* mm belongs to a dovetailed process */
>>>     #define MMF_TOPDOWN		31	/* mm searches top
>>> down by default */
>>>   #define MMF_TOPDOWN_MASK	(1 << MMF_TOPDOWN)
>
> Tried that on top of  v6.12.y-evl-rebase , it doesn't seem to fix the problem, neither for latmus nor for our application.
>
>> Note: this issue only affects kernels from v6.10 and on
>
> I can confirm that on v6.1.y-cip-evl-rebase I don't see that issue, even though MMF_DOVETAILED is 31 also there.

Quick check: does "evl check" run without any output (as expected)?

-- 
Philippe.