Testers wanted - MPSAFE watchdog for rge(4)

Index: if_rge.c
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/if_rge.c,v
retrieving revision 1.47
diff -u -p -r1.47 if_rge.c
--- if_rge.c	28 Jan 2026 06:15:37 -0000	1.47
+++ if_rge.c	1 Feb 2026 01:04:32 -0000
@@ -94,6 +94,11 @@ MCLGETL(struct rge_softc *sc __unused, i
 #endif
 #endif
 
+#ifndef RGE_WATCHDOG_TIMEOUT
+#define RGE_WATCHDOG_TIMEOUT 5
+#endif
+static int rge_watchdog_timeout = RGE_WATCHDOG_TIMEOUT;
+
 #ifdef RGE_DEBUG
 #define DPRINTF(x)	do { if (rge_debug > 0) printf x; } while (0)
 int rge_debug = 0;
@@ -109,7 +114,6 @@ static int	rge_encap(struct rge_softc *,
     struct mbuf *, int);
 static int	rge_ioctl(struct ifnet *, u_long, void *);
 static void	rge_start(struct ifnet *);
-static void	rge_watchdog(struct ifnet *);
 static int	rge_init(struct ifnet *);
 static void	rge_stop(struct ifnet *, int);
 static int	rge_ifmedia_upd(struct ifnet *);
@@ -120,6 +124,7 @@ static int	rge_newbuf(struct rge_queues 
 static int	rge_rx_list_init(struct rge_queues *);
 static void	rge_rx_list_fini(struct rge_queues *);
 static void	rge_tx_list_init(struct rge_queues *);
+/* static void	rge_tx_list_fini(struct rge_queues *); */
 static int	rge_rxeof(struct rge_softc *);
 static int	rge_txeof(struct rge_softc *);
 static int	rge_reset(struct rge_softc *);
@@ -171,9 +176,11 @@ static uint16_t	rge_read_phy(struct rge_
 static void	rge_write_phy_ocp(struct rge_softc *, uint16_t, uint16_t);
 static uint16_t	rge_read_phy_ocp(struct rge_softc *, uint16_t);
 static int	rge_get_link_status(struct rge_softc *);
-static void	rge_txstart(void *);
+static void	rge_txstart(struct rge_softc *);
 static void	rge_tick(void *);
 static void	rge_link_state(struct rge_softc *);
+static bool	rge_watchdog_tick(struct ifnet *);
+static void	rge_handle_reset_work(struct work *, void *);
 
 static const struct {
 	uint16_t reg;
@@ -367,6 +374,16 @@ rge_attach(device_t parent, device_t sel
 	if (rge_allocmem(sc))
 		return;
 
+	char wqname[MAXCOMLEN];
+	snprintf(wqname, sizeof(wqname), "%sReset", device_xname(sc->sc_dev));
+	int error = workqueue_create(&sc->sc_reset_wq, wqname,
+	    rge_handle_reset_work, sc, PRI_NONE, IPL_SOFTCLOCK,WQ_MPSAFE);
+	if (error) {
+		aprint_error_dev(sc->sc_dev,
+		    "unable to create reset workqueue\n");
+		return;
+	}
+
 	ifp = &sc->sc_ec.ec_if;
 	ifp->if_softc = sc;
 	strlcpy(ifp->if_xname, device_xname(sc->sc_dev), IFNAMSIZ);
@@ -378,7 +395,7 @@ rge_attach(device_t parent, device_t sel
 	ifp->if_stop = rge_stop;
 	ifp->if_start = rge_start;
 	ifp->if_init = rge_init;
-	ifp->if_watchdog = rge_watchdog;
+	ifp->if_watchdog = NULL;
 	IFQ_SET_MAXLEN(&ifp->if_snd, RGE_TX_LIST_CNT - 1);
 
 	ifp->if_capabilities = IFCAP_CSUM_IPv4_Rx |
@@ -388,7 +405,7 @@ rge_attach(device_t parent, device_t sel
 	sc->sc_ec.ec_capabilities |= ETHERCAP_VLAN_MTU;
 	sc->sc_ec.ec_capabilities |= ETHERCAP_VLAN_HWTAGGING;
 
-	callout_init(&sc->sc_timeout, CALLOUT_FLAGS);
+	callout_init(&sc->sc_timeout, CALLOUT_MPSAFE);
 	callout_setfunc(&sc->sc_timeout, rge_tick, sc);
 
 	command = pci_conf_read(pa->pa_pc, pa->pa_tag, PCI_COMMAND_STATUS_REG);
@@ -735,15 +752,67 @@ rge_start(struct ifnet *ifp)
 	rge_txstart(sc);
 }
 
-static void
-rge_watchdog(struct ifnet *ifp)
+static bool
+rge_watchdog_check(struct rge_softc * const sc)
 {
-	struct rge_softc *sc = ifp->if_softc;
 
-	device_printf(sc->sc_dev, "watchdog timeout\n");
+	KASSERT(mutex_owned(sc->sc_core_lock));
+
+	if (!sc->sc_tx_sending)
+		return true;
+
+	if (time_uptime - sc->sc_tx_lastsent <= rge_watchdog_timeout)
+		return true;
+
+	return false;
+}
+
+static bool
+rge_watchdog_tick(struct ifnet *ifp)
+{
+	struct rge_softc * const sc = ifp->if_softc;
+
+	KASSERT(mutex_owned(sc->sc_core_lock));
+
+	if (!sc->sc_trigger_reset && rge_watchdog_check(sc))
+		return true;
+
 	if_statinc(ifp, if_oerrors);
 
+	if (atomic_swap_uint(&sc->sc_reset_pending, 1) == 0)
+		workqueue_enqueue(sc->sc_reset_wq, &sc->sc_reset_work, NULL);
+
+	return false;
+}
+
+/*
+ * Perform an interface watchdog reset.
+ */
+static void
+rge_handle_reset_work(struct work *work, void *arg)
+{
+	struct rge_softc * const sc = arg;
+	struct ifnet * const ifp = &sc->sc_ec.ec_if;
+
+	printf("%s: watchdog timeout -- resetting\n", ifp->if_xname);
+
+	/* Don't want ioctl operations to happen */
+	IFNET_LOCK(ifp);
+
+	/* reset the interface. */
 	rge_init(ifp);
+
+	IFNET_UNLOCK(ifp);
+	/*
+	 * There are still some upper layer processing which call
+	 * ifp->if_start(). e.g. ALTQ or one CPU system
+	 */
+
+	/* Try to get more packets going. */
+	ifp->if_start(ifp);
+
+	atomic_store_relaxed(&sc->sc_reset_pending, 0);
+
 }
 
 static int
@@ -3804,9 +3873,8 @@ rge_get_link_status(struct rge_softc *sc
 }
 
 static void
-rge_txstart(void *arg)
+rge_txstart(struct rge_softc *sc)
 {
-	struct rge_softc *sc = arg;
 
 	RGE_WRITE_2(sc, RGE_TXSTART, RGE_TXSTART_START);
 }
@@ -3815,13 +3883,21 @@ static void
 rge_tick(void *arg)
 {
 	struct rge_softc *sc = arg;
-	int s;
 
-	s = splnet();
+	mutex_enter(sc->sc_core_lock);
+	if (sc->sc_stopping) {
+		mutex_exit(sc->sc_core_lock);
+		return;
+	}
+
 	rge_link_state(sc);
-	splx(s);
 
-	callout_schedule(&sc->sc_timeout, hz);
+	struct ifnet * const ifp = &sc->sc_ec.ec_if;
+	const bool ok = rge_watchdog_tick(ifp);
+	if (ok)
+		callout_schedule(&sc->sc_timeout, hz);
+
+	mutex_exit(sc->sc_core_lock);
 }
 
 static void
Index: if_rgereg.h
===================================================================
RCS file: /cvsroot/src/sys/dev/pci/if_rgereg.h,v
retrieving revision 1.18
diff -u -p -r1.18 if_rgereg.h
--- if_rgereg.h	28 Jan 2026 06:15:37 -0000	1.18
+++ if_rgereg.h	1 Feb 2026 01:04:34 -0000
@@ -424,6 +424,18 @@ struct rge_softc {
 	int			rge_timerintr;
 #define RGE_IMTYPE_NONE		0
 #define RGE_IMTYPE_SIM		1
+
+	kmutex_t		*sc_core_lock;
+	time_t			sc_tx_lastsent;
+	bool			sc_tx_sending;
+
+	struct workqueue 	*sc_reset_wq;
+	struct work		sc_reset_work;
+	volatile unsigned	sc_reset_pending;
+	bool			sc_stopping;
+
+	bool			sc_trigger_reset;
+
 };
 
 /*