On Mon, Sep 23, 2024 at 12:05:53PM +0200, Oleg Nesterov wrote: > On 09/23, Jiri Olsa wrote: > > > > change below should do what you proposed originally > > LGTM, just one nit below. > > But I guess you need to do this on top of bpf/bpf.git, Andrii has already > applied your series. that seems confusing but looks like just that one fix with the commit link in [1] was applied [1] https://lore.kernel.org/bpf/172708047825.3261420.5126267811201364070.git-patchwork-notify@xxxxxxxxxx/T/#mb065649b5ab8f7ea5b03c215bdc6555a0b76c0d7 > > And to remind, 02/14 must be fixed in any case unless I am totally confused, > handler_chain() can leak return_instance. yep it was missing kfree, but it's not needed in this new version > > > also on top of that.. I discussed with Andrii the possibility of dropping > > the UPROBE_HANDLER_IWANTMYCOOKIE completely and setup cookie for any consumer > > that has both 'handler' and 'ret_handler' defined, wdyt? > > Up to you. As I said from the very beginning I won't insist on _IWANTMYCOOKIE. ok > > > list_for_each_entry_srcu(uc, &uprobe->consumers, cons_node, > > srcu_read_lock_held(&uprobes_srcu)) { > > + ric = return_consumer_find(ri, &ric_idx, uc->id); > > if (uc->ret_handler) > > - uc->ret_handler(uc, ri->func, regs); > > + uc->ret_handler(uc, ri->func, regs, ric ? &ric->cookie : NULL); > > } > > srcu_read_unlock(&uprobes_srcu, srcu_idx); > > return_consumer_find() makes no sense if !uc->ret_handler, can you move > > ric = return_consumer_find(ri, &ric_idx, uc->id); > > into the "if (uc->ret_handler)" block? ok, will move that thanks, jirka
