Hi, On 6/5/2025 1:55 AM, Alexei Starovoitov wrote: > On Mon, Jun 2, 2025 at 9:01 PM Hou Tao <houtao@xxxxxxxxxxxxxxx> wrote: >>>> Maybe adding a >>>> bit flag to the pointer of the element to indicate that it has been >>>> destroyed is OK ? Will try it in the next revision. >>> Not sure what you have in mind. >> Er. I was just wondering that if it is possible to the bit 0 of the >> element pointer to indicate that the element has been destroyed, >> therefore, the second invocation of dtor can be avoided. > but the element was deleted from htab bucket, > so what pointer will you tag with extra bit? Sorry for the late reply. Had a lot of work recently. I was thinking that when the pointer has been freed through bpf_mem_cache_free(), the pointer would be tracked in llist_node. It would be possible to reuse the bit 0 of ->first and ->next of the lockless list to indicate whether the pointed pointer has already been destroyed or not. After the second thought, that would need to duplicate each lockless list API and may not worth it.
