On Fri, Oct 28, 2022 at 07:14:16PM +0300, Leon Romanovsky wrote: > On Fri, Oct 28, 2022 at 01:00:27PM -0300, Jason Gunthorpe wrote: > > On Fri, Oct 28, 2022 at 09:58:56AM +0300, Leon Romanovsky wrote: > > > ppc64_defconfig) produced this warning: > > > > > > WARNING: modpost: drivers/infiniband/core/ib_core.o: section mismatch in reference: .init_module (section: .init.text) -> .nldev_exit (section: .exit.text) > > > > > > Fix it by removing __init/__exit markers as nldev is part of ib_core.ko > > > and as such doesn't require any special notations for entry/exit functions. > > > > This isn't what the problem is, the patch Stephen reported: > > > > commit ad9394a3da33995dff828dbfd4540421e535bec9 (ko-rdma/for-rc) > > Author: Chen Zhongjin <chenzhongjin@xxxxxxxxxx> > > Date: Tue Oct 25 10:41:46 2022 +0800 > > > > RDMA/core: Fix null-ptr-deref in ib_core_cleanup() > > > > Adds a call to an __exit function from an __init function: > > > > @@ -2815,10 +2815,18 @@ static int __init ib_core_init(void) > > > > +err_parent: > > + rdma_nl_unregister(RDMA_NL_LS); > > + nldev_exit(); > > + unregister_pernet_device(&rdma_dev_net_ops); > > > > Which is not allowed > > > > All that is required is to drop the __exit from nldev_exit, > > This is why I dropped both __exit and __init. I see no value in keeping > __init, without __exit. __init works just fine, it will cause the code to be unload after the module registration is compelted __exit allows the code to be removed if the module is compiled in and the __exit is never called Jason