Michal Hocko <mhocko at kernel.org> writes:
> On Thu 16-06-16 12:30:16, Vitaly Kuznetsov wrote:
>> Christoph Hellwig <hch at infradead.org> writes:
>>
>> > On Thu, Jun 16, 2016 at 11:22:46AM +0200, Vitaly Kuznetsov wrote:
>> >> _count -> _refcount rename in commit 0139aa7b7fa12 ("mm: rename _count,
>> >> field of the struct page, to _refcount") broke kdump. makedumpfile(8) does
>> >> stuff like READ_MEMBER_OFFSET("page._count", page._count) and fails. While
>> >> it is definitely possible to fix this particular tool I'm not sure about
>> >> other tools which might be doing the same.
>> >>
>> >> I suggest we remember the "we don't break userspace" rule and revert for
>> >> 4.7 while it's not too late.
>> >
>> > Err, sorry - this is not "userspace". It's crazy crap digging into
>> > kernel internal structure.
>> >
>> > The rename was absolutely useful, so fix up your stinking pike in kdump.
>>
>> Ok, sure, I'll send a patch to it. I was worried about other tools out
>> there which e.g. inspect /proc/vmcore. As it is something we support
>> some conservatism around it is justified.
>
> struct page layout as some others that such a tool might depend on has
> changes several times in the past so I fail to see how is it any
> different this time.
IMO this time the change doesn't give us any advantage, it was just a
rename.
> struct page is nothing the userspace should depend on.
True but at least makedumpfile(8) is special and even if it's a 'crazy
crap digging into ...' we could avoid breaking it for no technical
reason.
--
Vitaly
