Re: [PATCH 1/3] mm/vma: miss to restore vmi.index on expansion failure

[Wei Yang <richard.weiyang@gmail.com>]

On Fri, Oct 25, 2024 at 10:01:36AM +0100, Lorenzo Stoakes wrote:
>On Fri, Oct 25, 2024 at 08:54:09AM +0000, Wei Yang wrote:
>> On Fri, Oct 25, 2024 at 09:51:12AM +0100, Lorenzo Stoakes wrote:
>> >On Fri, Oct 25, 2024 at 08:49:19AM +0000, Wei Yang wrote:
>> >> On Fri, Oct 25, 2024 at 09:40:25AM +0100, Lorenzo Stoakes wrote:
>> >> >On Fri, Oct 25, 2024 at 08:32:27AM +0000, Wei Yang wrote:
>> >> >
>> >> >>
>> >> >> But I have a question on your introduction of VMG_FLAG_JUST_EXPAND.
>> >> >>
>> >> >> Lets say just_expand is true and can_merge_left is true. Now we will adjust
>> >> >> vmg->start/vma/pgoff in if (can_merge_left). If we fail expansion, we won't
>> >> >> restore vmg->vma/start/pgoff, since just_expand is true.
>> >> >>
>> >> >> Is this what you expect?
>> >> >
>> >> >Yes, I explicitly wrote it to do that so it'd be a bit odd if I didn't
>> >> >realise :)
>> >> >
>> >> >Actually at this point, I don't think we need a follow up patch, sorry.
>> >> >
>> >> >As I think perhaps I will make this change as part of an existing series
>> >> >where I am reworking mmap_region(), since this is the only place where it
>> >> >matters, and it would make everything a hell of a lot clearer.
>> >> >
>> >> >Thanks for pointing this out, it's very useful (and an embarrassing
>> >> >oversight on my part...!), but I think it'd be better reworked this way.
>> >> >
>> >> >Thanks!
>> >>
>> >> Ok, for now I would just remove these two lines with the change log you
>> >> suggested.
>> >
>> >NO! Sorry I've not been clear - don't send any series.
>> >
>> >I am going to make a change that eliminates the need for your change (sorry, but
>> >in discussing this I've realised that's the best way forward).
>> >
>>
>> Sure, go ahead.
>
>Thanks, apologies I know it sucks, to be clear I very much appreciate your
>input here!
>

Ha, np. Glad to talk with you.

>For more detail - we are refactoring how the '2nd attempt' at a merge
>works, also avoiding any abuse of vmg-> fields to keep track of start/end
>of a range, so it becomes better to just open-code 'reset' of these fields
>precisely at the point we need them.
>

I am not sure we mention the same thing.

One confusion during code reading is on the vmg->start/end and
vmg->vmi->mas->index/last.

For many times I lost who is who...

>Do please let us know if you notice any other silly mistakes like this :>)
>

Let me clear my glasses :-)

>Thanks, Lorenzo
>
>>
>> >>
>> >> --
>> >> Wei Yang
>> >> Help you, Help me
>> >>
>>
>> --
>> Wei Yang
>> Help you, Help me
>>

-- 
Wei Yang
Help you, Help me