port-xen/59745: guest network hosed, host says: xvif6i0 GNTTABOP_copy[0] Rx -3

[campbell+netbsd%mumble.net@localhost]

>Number:         59745
>Category:       port-xen
>Synopsis:       guest network hosed, host says: xvif6i0 GNTTABOP_copy[0] Rx -3
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    port-xen-maintainer
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Nov 03 01:25:00 +0000 2025
>Originator:     Taylor R Campbell
>Release:        10.1_STABLE
>Organization:
The GrantBSD Foundabop
>Environment:
NetBSD cole.netbsd.de 10.1_STABLE NetBSD 10.1_STABLE (TNF_XENSERVER_PVH) #1: Sat Oct  4 22:20:13 UTC 2025  spz%franklin.NetBSD.org@localhost:/home/netbsd/10/amd64/obj/sys/arch/amd64/compile/TNF_XENSERVER_PVH amd64
>Description:

	In two incidents recently, we've had a guest on TNF
	infrastructure become unreachable over the network -- no
	response to pings or TCP connections, nothing logged in
	npflog0, system mainly idle.

	The host's dmesg is full of messages like this:

[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3
[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3
[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3
[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3
[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3
[ 2451339.802864] xvif6i0 GNTTABOP_copy[0] Rx -3

	That's this message:

    638 	/*
    639 	 * Copy the data and ack it. Delaying it until the mbuf is
    640 	 * freed will stall transmit.
    641 	 */
    642 	if (HYPERVISOR_grant_table_op(GNTTABOP_copy, gop, copycnt) != 0) {
    643 		printf("%s: GNTTABOP_copy %s failed", ifp->if_xname, dir);
    644 		return EINVAL;
    645 	}
    646 
    647 	for (int i = 0; i < copycnt; i++) {
    648 		if (gop->status != GNTST_okay) {
    649 			printf("%s GNTTABOP_copy[%d] %s %d\n",
    650 			    ifp->if_xname, i, dir, gop->status);
    651 			return EINVAL;
    652 		}
    653 	}

	https://nxr.NetBSD.org/xref/src/sys/arch/xen/xen/xennetback_xenbus.c?r=1.108.4.3#649

	Restarting the guest makes the problem go away, for a little
	while.

	Internal admins ticket references:
	https://rt.NetBSD.org/Ticket/Display.html?id=377840
	https://rt.NetBSD.org/Ticket/Display.html?id=372971

>How-To-Repeat:

	No idea, maybe run a network-oriented guest for a while.

>Fix:

	Yes, please!