Re: [PATCH v7 1/3] kho: make kho_scratch_overlap usable outside debugging

[Pratyush Yadav]

On Tue, Apr 07 2026, Pasha Tatashin wrote:

> On Tue, Apr 7, 2026 at 6:55 AM Pratyush Yadav <pratyush@xxxxxxxxxx> wrote:
>>
>> On Wed, Mar 18 2026, Mike Rapoport wrote:
>>
>> > Hi Michal,
>> >
>> > On Tue, Mar 17, 2026 at 03:15:32PM +0100, Michal Clapinski wrote:
>> >> Also return false if kho_scratch is NULL.
>> >>
>> [...]
>> >> diff --git a/kernel/liveupdate/Makefile b/kernel/liveupdate/Makefile
>> >> index d2f779cbe279..dc352839ccf0 100644
>> >> --- a/kernel/liveupdate/Makefile
>> >> +++ b/kernel/liveupdate/Makefile
>> >> @@ -7,7 +7,6 @@ luo-y :=                                                             \
>> >>              luo_session.o
>> >>
>> >>  obj-$(CONFIG_KEXEC_HANDOVER)                += kexec_handover.o
>> >> -obj-$(CONFIG_KEXEC_HANDOVER_DEBUG)  += kexec_handover_debug.o
>> >>  obj-$(CONFIG_KEXEC_HANDOVER_DEBUGFS)        += kexec_handover_debugfs.o
>> >>
>> >>  obj-$(CONFIG_LIVEUPDATE)            += luo.o
>> >> diff --git a/kernel/liveupdate/kexec_handover.c b/kernel/liveupdate/kexec_handover.c
>> >> index 532f455c5d4f..c9b982372d6e 100644
>> >> --- a/kernel/liveupdate/kexec_handover.c
>> >> +++ b/kernel/liveupdate/kexec_handover.c
>> >> @@ -820,7 +820,8 @@ int kho_preserve_folio(struct folio *folio)
>> >>      const unsigned long pfn = folio_pfn(folio);
>> >>      const unsigned int order = folio_order(folio);
>> >>
>> >> -    if (WARN_ON(kho_scratch_overlap(pfn << PAGE_SHIFT, PAGE_SIZE << order)))
>> >> +    if (WARN_ON(kho_scratch_overlap_debug(pfn << PAGE_SHIFT,
>> >> +                                          PAGE_SIZE << order)))
>> >
>> > Can't say I'm fond of kho_scratch_overlap_debug(). How about we make it
>> >
>> >       if (IS_ENABLED(CONFIG_KEXEC_HANDOVER_DEBUG) &&
>> >           WARN_ON(kho_scratch_overlap(...))
>>
>> +1. And we can get rid of kexec_handover_debug.c, for now at least. We
>> can add it back when we have something else to put in there.
>
> Are you proposing moving kho_scratch_overlap() into kexec_handover.c?
> That would make it uglier to have #ifdefs in the C file. If you mean
> removing this function entirely, I think that is too dangerous because
> we have already had a memory corruption issue [1] that was challenging
> to root cause, and having this simple check prevents this from
> occurring going forward. The problem is that changes to defences such
> as kfence, kasan, and asi are happening outside of the core KHO code,
> and it is very easy to miss when something unexpectedly causes a
> preservation from the scratch area, as we have seen this with kfence.
> Worst of all, some of those mitigations use randomized or sampling
> approaches and might not be reproducible on every try, so having a
> CONFIG that tests it every time in a debug build is the only solid
> defense against that.

I think you miss the context here. This patchset uses
kho_scratch_overlap() during MM init to set the migrate type of
pageblocks. So it will no longer be gated by
CONFIG_KEXEC_HANDOVER_DEBUG, but by CONFIG_KEXEC_HANDOVER instead. So
there is no need for any #ifdefs. All we need to change is to have the
debug checks gated with a IS_ENABLED(CONFIG_KEXEC_HANDOVER_DEBUG).

So the function stays around, and so do the debug checks. Since core KHO
now uses this function, we just move it out to the main file. And since
kexec_handover_debug.c has nothing else, we can delete it for now.

Anyway, based on the discussion it looks like people want to ask
memblock directly and not use kho_scratch_overlap(), so the next version
might not have this patch at all.

>
> Pasha
>
> [1] https://lore.kernel.org/all/20251021000852.2924827-1-pasha.tatashin@xxxxxxxxxx
>
>>
>> >
>> >>              return -EINVAL;
>> >>
>> >>      return kho_radix_add_page(tree, pfn, order);
>> [...]
>>
>> --
>> Regards,
>> Pratyush Yadav

-- 
Regards,
Pratyush Yadav