Re: [PATCH] x86/MCE/AMD: Decrement threshold_bank refcount when removing threshold blocks

From: Borislav Petkov
Date: Sat Aug 13 2022 - 06:10:32 EST


On Fri, Aug 12, 2022 at 11:14:44PM +0200, Mateusz Jończyk wrote:
> Shouldn't there be "kobject_put(&pos->kobj)" here instead?

Yes, it should.

> Also, it seems to me that "kobject_put(b->kobj);" before the loop
> may be relocated after the loop - so that the refcounts on the child
> objects are decreased first, then the refcount on the parent object.

Yes, I guess we can do that.

> Additionally, shouldn't there be a call to
> "kobject_put(&b->blocks->kobj);" in __threshold_remove_blocks()?

Makes sense, we do

kobject_add(&b->blocks->kobj, ...

in __threshold_add_blocks().

> From what I understand, b->blocks is a list head, so we need to
> decrease the refcount on it too.

Not list_heads - we modify the refcount of kobjects. See what the arg of
kobject_put() is.

> After these changes, the __threshold_remove_blocks() function looks
> very similar to deallocate_threshold_blocks() function just above it.

Yes, minus the list_del(&pos->miscj); But that can be made conditional
with a bool arg to deallocate_threshold_blocks() and then remove
__threshold_remove_blocks().

Care to take Yazen's patch, fix it up, test it thoroughly (you should
enable KASAN to catch any potential memory leaks) and send it?

Thx.

--
Regards/Gruss,
Boris.

https://people.kernel.org/tglx/notes-about-netiquette