On Wed, Jul 31, 2019 at 7:22 PM Chuhong Yuan <hsleste...@gmail.com> wrote: > > Michael Chan <michael.c...@broadcom.com> 于2019年8月1日周四 上午1:58写道: > > > > On Wed, Jul 31, 2019 at 5:22 AM Chuhong Yuan <hsleste...@gmail.com> wrote: > > > > > static void cnic_ctx_wr(struct cnic_dev *dev, u32 cid_addr, u32 off, u32 > > > val) > > > @@ -494,7 +494,7 @@ int cnic_register_driver(int ulp_type, struct > > > cnic_ulp_ops *ulp_ops) > > > } > > > read_unlock(&cnic_dev_lock); > > > > > > - atomic_set(&ulp_ops->ref_count, 0); > > > + refcount_set(&ulp_ops->ref_count, 0); > > > rcu_assign_pointer(cnic_ulp_tbl[ulp_type], ulp_ops); > > > mutex_unlock(&cnic_lock); > > > > > > > Willem's comment applies here too. The driver needs to be modified to > > count from 1 to use refcount_t. > > > > Thanks. > > I have revised this problem but find the other two refcounts - > cnic_dev::ref_count > and cnic_sock::ref_count have no set. > I am not sure where to initialize them to 1. > > Besides, should ulp_ops->ref_count be set to 0 when unregistered?
I will send a patch to fix up the initialization of all the atomic ref counts. After that, you can add your patch to convert to refcount_t.