The NFS subsystem is not MP-safe yet. Take this into account in the NFS server socket upcall by locking the kernel.
This might help with the NFS server hanging that was seen recently as a result of the now-reverted selwakeup()-to-KNOTE() conversion. Unfortunately, I have not been able to confirm this myself. OK? Index: nfs/nfs_socket.c =================================================================== RCS file: src/sys/nfs/nfs_socket.c,v retrieving revision 1.140 diff -u -p -r1.140 nfs_socket.c --- nfs/nfs_socket.c 17 Mar 2022 14:23:34 -0000 1.140 +++ nfs/nfs_socket.c 13 May 2022 15:38:48 -0000 @@ -1561,8 +1561,10 @@ nfsrv_rcv(struct socket *so, caddr_t arg struct uio auio; int flags, error; + KERNEL_LOCK(); + if ((slp->ns_flag & SLP_VALID) == 0) - return; + goto out; /* Defer soreceive() to an nfsd. */ if (waitflag == M_DONTWAIT) { @@ -1644,6 +1646,9 @@ dorecs: if (waitflag == M_DONTWAIT && (slp->ns_rec || (slp->ns_flag & (SLP_NEEDQ | SLP_DISCONN)))) nfsrv_wakenfsd(slp); + +out: + KERNEL_UNLOCK(); } /*