On Thu, Sep 01, 2022 at 04:47:11PM +0800, Vlastimil Babka wrote:
> On 9/1/22 03:26, Feng Tang wrote:
> > On Thu, Sep 01, 2022 at 12:16:17AM +0800, Vlastimil Babka wrote:
> >> On 8/31/22 16:21, Marco Elver wrote:
> >> > On Wed, 31 Aug 2022 at 16:04, Hyeonggon Yoo <42.hyeyoo@xxxxxxxxx> wrote:
> >> >
> >> >> Maybe you can include those functions too?
> >> >>
> >> >> - __kmem_cache_alloc_node
> >> >> - kmalloc_[node_]trace, kmalloc_large[_node]
> >> >
> >> > This is only required if they are allocator "root" functions when
> >> > entering allocator code (or may be tail called by a allocator "root"
> >> > function). Because get_stack_skipnr() looks for one of the listed
> >> > function prefixes in the whole stack trace.
> >> >
> >> > The reason __kmem_cache_free() is now required is because it is tail
> >> > called by kfree() which disappears from the stack trace if the
> >> > compiler does tail-call-optimization.
> >>
> >> I checked and I have this jmp tail call, yet all test pass here.
> >> But I assume the right commit to amend is
> >> 05a1c2e50809 ("mm/sl[au]b: generalize kmalloc subsystem")
> >>
> >> Could you Feng maybe verify that that commit is the first that fails the
> >> tests, and parent commit of that is OK? Thanks.
> >
> > Yes, 05a1c2e50809 is the first commit that I saw the 4 kfence failed
> > kunit cases.
>
> Thanks, squashed your patch there and pushed new for-next.
Thanks! Just re-pulled slab tree's 'for-next' branch and the error
can't be reproduced with it.
- Feng
