Re: [PATCH v10 6/6] selftests/mm: add hwpoison-panic destructive test

On 7/29/26 10:53, Breno Leitao wrote:
> On Wed, Jul 29, 2026 at 09:25:16AM +0200, David Hildenbrand (Arm) wrote:
>> On 7/29/26 00:07, Breno Leitao wrote:
>>> I have a draft of the selftest in C already and I will give it more bake
>>> time, and I will submit it until EOW.
>>
>> IMHO there is no need to rush this test here, we could merge it after the main work.
>>
>> We neglected our selftests quite a bit in the past, and that is biting us now.
>>
>> That's why I push back hard now (sorry) against stuff that is different than the
>> rest and not easy to maintain.
>>
>> I know, hwpoison-panic is a destructive test, but I really don't think we should
>> just merge mostly unreviewed bash tests that will just bitrot.
>>
>> Which makes me wonder now: should this actually be a selftest? It's nothing one
>> would just want to run in a CI. It will crash your machine.
>>
>> Maybe this should just be a tool in tools/mm? I really think so. So having a
>> nice C-level hwpoison-panic.c in tools/mm might be the way to go.
> 
> Agreed. In v6 [0] I actually removed the selftest entirely, since the value
> wasn't clear and it seemed like an ugly duckling.
> 
> It seems tools/mm would be a good fit, and make a better home for this
> type of "test".
> 
> Given more changes coming in memory failure (at least from my side
> [1][2]), I think it would be a good idea to create a subdirectory
> (tools/mm/memory-failure) and keep these semi-destructive tools in there. Is
> this the right direction?

Makes sense to me.

> 
>> So let's rather make this test well maintainable and have it properly reviewed
>> than rushing something suboptimal.
> 
> Sounds good. I'll move this to tools/mm and resubmit. Given Andrew has
> removed the selftest from mm-unstable while keeping the rest of the
> series (which is fully reviewed and frictionless), so I'll focus on the
> test tool now and send it separately.

Avoid bash scripts, please :)

-- 
Cheers,

David