sync with OpenBSD -current

regress/usr.sbin/bgpd/integrationtests/Makefile

@@ -1,4 +1,4 @@
-# $OpenBSD: Makefile,v 1.26 2025/01/13 14:18:07 claudio Exp $
+# $OpenBSD: Makefile,v 1.27 2025/01/14 13:15:18 claudio Exp $
 
 REGRESS_TARGETS	= 	network_statement md5 ovs capa policy pftable \
 			mrt maxprefix maxprefixout maxcomm maxattr \
@@ -51,6 +51,9 @@ l3vpn:
 ixp:
 	${SUDO} ksh ${.CURDIR}/$@.sh ${BGPD} ${.CURDIR} 11 12 pair11 pair12
 
+addpath:
+	${SUDO} ksh ${.CURDIR}/$@.sh ${BGPD} ${.CURDIR} 11 12 pair11 pair12
+
 lladdr:
 	${SUDO} ksh ${.CURDIR}/$@.sh ${BGPD} ${.CURDIR} 11 12 pair11 pair12
 

regress/usr.sbin/bgpd/integrationtests/addpath.rdomain2.ok

@@ -0,0 +1,31 @@
+flags: * = Valid, > = Selected, I = via IBGP, A = Announced,
+       S = Stale, E = Error, F = Filtered
+origin validation state: N = not-found, V = valid, ! = invalid
+aspa validation state: ? = unknown, V = valid, ! = invalid
+origin: i = IGP, e = EGP, ? = Incomplete
+
+flags  vs destination          gateway          lpref   med aspath origin
+*>    N-? 2.0.1.0/24           192.0.2.21        100     2 2 i
+*>    N-? 2.0.2.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.2.0/24           192.0.2.31        100     3 3 i
+*>    N-? 2.0.3.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.3.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.3.0/24           192.0.2.41        100     4 4 i
+*>    N-? 2.0.4.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.4.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.4.0/24           192.0.2.41        100     4 4 i
+*>    N-? 2.0.5.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.5.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.5.0/24           192.0.2.41        100     4 4 i
+*>    N-? 2.0.6.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.6.0/24           192.0.2.41        100     4 4 i
+*     N-? 2.0.6.0/24           192.0.2.51        100     5 5 i
+*>    N-? 2.0.7.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.7.0/24           192.0.2.41        100     4 4 i
+*     N-? 2.0.7.0/24           192.0.2.51        100     5 5 i
+*>    N-? 2.0.8.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.8.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.8.0/24           192.0.2.51        100     5 5 i
+*>    N-? 2.0.9.0/24           192.0.2.21        100     2 2 i
+*     N-? 2.0.9.0/24           192.0.2.31        100     3 3 i
+*     N-? 2.0.9.0/24           192.0.2.41        100     4 4 i

regress/usr.sbin/bgpd/integrationtests/addpath.sh

@@ -0,0 +1,100 @@
+#!/bin/ksh
+#	$OpenBSD: addpath.sh,v 1.1 2025/01/14 13:15:18 claudio Exp $
+
+set -e
+
+BGPD=$1
+BGPDCONFIGDIR=$2
+RDOMAIN1=$3
+RDOMAIN2=$4
+PAIR1=$5
+PAIR2=$6
+
+RDOMAINS="${RDOMAIN1} ${RDOMAIN2}"
+PAIRS="${PAIR1} ${PAIR2}"
+PAIR1IP=192.0.2.2
+PAIR2IP=192.0.2.11
+PAIR2IP2=192.0.2.21
+PAIR2IP3=192.0.2.31
+PAIR2IP4=192.0.2.41
+PAIR2IP5=192.0.2.51
+
+error_notify() {
+	echo cleanup
+	pkill -T ${RDOMAIN1} bgpd || true
+	pkill -T ${RDOMAIN2} bgpd || true
+	sleep 1
+	ifconfig ${PAIR2} destroy || true
+	ifconfig ${PAIR1} destroy || true
+	route -qn -T ${RDOMAIN1} flush || true
+	route -qn -T ${RDOMAIN2} flush || true
+	ifconfig lo${RDOMAIN1} destroy || true
+	ifconfig lo${RDOMAIN2} destroy || true
+	if [ $1 -ne 0 ]; then
+		echo FAILED
+		exit 1
+	else
+		echo SUCCESS
+	fi
+}
+
+if [ "$(id -u)" -ne 0 ]; then 
+	echo need root privileges >&2
+	exit 1
+fi
+
+. "${BGPDCONFIGDIR}/util.sh"
+
+trap 'error_notify $?' EXIT
+
+echo check if rdomains are busy
+for n in ${RDOMAINS}; do
+	if /sbin/ifconfig | grep -v "^lo${n}:" | grep " rdomain ${n} "; then
+		echo routing domain ${n} is already used >&2
+		exit 1
+	fi
+done
+
+echo check if interfaces are busy
+for n in ${PAIRS}; do
+	/sbin/ifconfig "${n}" >/dev/null 2>&1 && \
+	    ( echo interface ${n} is already used >&2; exit 1 )
+done
+
+set -x
+
+echo setup
+ifconfig ${PAIR1} rdomain ${RDOMAIN1} ${PAIR1IP}/24 up
+ifconfig ${PAIR2} rdomain ${RDOMAIN2} ${PAIR2IP}/24 up
+ifconfig ${PAIR2} alias ${PAIR2IP2}/32
+ifconfig ${PAIR2} alias ${PAIR2IP3}/32
+ifconfig ${PAIR2} alias ${PAIR2IP4}/32
+ifconfig ${PAIR2} alias ${PAIR2IP5}/32
+ifconfig ${PAIR1} patch ${PAIR2}
+ifconfig lo${RDOMAIN1} inet 127.0.0.1/8
+ifconfig lo${RDOMAIN2} inet 127.0.0.1/8
+
+echo run bgpds
+route -T ${RDOMAIN1} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain1.conf
+sleep 2
+route -T ${RDOMAIN2} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain2_1.conf
+route -T ${RDOMAIN2} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain2_2.conf
+route -T ${RDOMAIN2} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain2_3.conf
+route -T ${RDOMAIN2} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain2_4.conf
+route -T ${RDOMAIN2} exec ${BGPD} \
+	-v -f ${BGPDCONFIGDIR}/bgpd.addpath.rdomain2_5.conf
+
+sleep 7
+
+route -T ${RDOMAIN1} exec bgpctl show
+route -T ${RDOMAIN2} exec bgpctl show rib | tee addpath.rdomain2.out
+
+diff -u ${BGPDCONFIGDIR}/addpath.rdomain2.ok addpath.rdomain2.out
+echo OK
+
+exit 0

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain1.conf

@@ -0,0 +1,51 @@
+# built by ARouteServer
+AS 999
+router-id 192.0.2.2
+
+fib-update no
+log updates
+
+nexthop qualify via default
+
+rde evaluate all
+
+# ---------------------------------------------------------
+# MEMBERS
+
+group "clients" {
+	transparent-as yes
+
+	neighbor 192.0.2.11 {
+		remote-as 1
+		descr "AS1_1 client"
+
+		announce add-path send best plus 2
+	}
+
+	neighbor 192.0.2.21 {
+		remote-as 2
+		descr "AS2_1 client"
+		set med +2
+	}
+
+	neighbor 192.0.2.31 {
+		remote-as 3
+		descr "AS3_1 client"
+		set med +3
+	}
+
+	neighbor 192.0.2.41 {
+		remote-as 4
+		descr "AS4_1 client"
+		set med +4
+	}
+
+	neighbor 192.0.2.51 {
+		remote-as 5
+		descr "AS5_1 client"
+		set med +5
+	}
+}
+
+allow from any
+allow to any

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain2_1.conf

@@ -0,0 +1,17 @@
+AS 1
+router-id 192.0.2.11
+listen on 192.0.2.11
+fib-update no
+
+rde med compare always
+
+neighbor 192.0.2.2 {
+	remote-as 999
+	local-address 192.0.2.11
+	enforce neighbor-as no
+
+	announce add-path recv yes
+}
+
+allow from any
+allow to any

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain2_2.conf

@@ -0,0 +1,24 @@
+AS 2
+router-id 192.0.2.21
+listen on 192.0.2.21
+fib-update no
+socket "/var/run/bgpd.sock.12_2"
+
+network 2.0.1.0/24
+network 2.0.2.0/24
+network 2.0.3.0/24
+network 2.0.4.0/24
+network 2.0.5.0/24
+network 2.0.6.0/24 set prepend-self 2
+network 2.0.7.0/24
+network 2.0.8.0/24
+network 2.0.9.0/24
+
+neighbor 192.0.2.2 {
+	remote-as 999
+	local-address 192.0.2.21
+	enforce neighbor-as no
+}
+
+deny from any
+allow to any

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain2_3.conf

@@ -0,0 +1,23 @@
+AS 3
+router-id 192.0.2.31
+listen on 192.0.2.31
+fib-update no
+socket "/var/run/bgpd.sock.12_3"
+
+network 2.0.2.0/24
+network 2.0.3.0/24
+network 2.0.4.0/24
+network 2.0.5.0/24
+network 2.0.6.0/24
+network 2.0.7.0/24 set prepend-self 2
+network 2.0.8.0/24
+network 2.0.9.0/24
+
+neighbor 192.0.2.2 {
+	remote-as 999
+	local-address 192.0.2.31
+	enforce neighbor-as no
+}
+
+deny from any
+allow to any

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain2_4.conf

@@ -0,0 +1,22 @@
+AS 4
+router-id 192.0.2.41
+listen on 192.0.2.41
+fib-update no
+socket "/var/run/bgpd.sock.12_4"
+
+network 2.0.3.0/24
+network 2.0.4.0/24
+network 2.0.5.0/24
+network 2.0.6.0/24
+network 2.0.7.0/24
+network 2.0.8.0/24 set prepend-self 2
+network 2.0.9.0/24
+
+neighbor 192.0.2.2 {
+	remote-as 999
+	local-address 192.0.2.41
+	enforce neighbor-as no
+}
+
+deny from any
+allow to any

regress/usr.sbin/bgpd/integrationtests/bgpd.addpath.rdomain2_5.conf

@@ -0,0 +1,21 @@
+AS 5
+router-id 192.0.2.51
+listen on 192.0.2.51
+fib-update no
+socket "/var/run/bgpd.sock.12_5"
+
+network 2.0.4.0/24
+network 2.0.5.0/24
+network 2.0.6.0/24
+network 2.0.7.0/24
+network 2.0.8.0/24
+network 2.0.9.0/24 set prepend-self 2
+
+neighbor 192.0.2.2 {
+	remote-as 999
+	local-address 192.0.2.51
+	enforce neighbor-as no
+}
+
+deny from any
+allow to any

sys/dev/fdt/virtio_mmio.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: virtio_mmio.c,v 1.21 2024/12/20 22:18:27 sf Exp $	*/
+/*	$OpenBSD: virtio_mmio.c,v 1.23 2025/01/14 14:28:38 sf Exp $	*/
 /*	$NetBSD: virtio.c,v 1.3 2011/11/02 23:05:52 njoly Exp $	*/
 
 /*
@@ -105,6 +105,8 @@ int		virtio_mmio_negotiate_features(struct virtio_softc *,
     const struct virtio_feature_name *);
 int		virtio_mmio_intr(void *);
 void		virtio_mmio_intr_barrier(struct virtio_softc *);
+int		virtio_mmio_intr_establish(struct virtio_softc *, struct virtio_attach_args *,
+    int, struct cpu_info *, int (*)(void *), void *);
 
 struct virtio_mmio_softc {
 	struct virtio_softc	sc_sc;
@@ -122,7 +124,7 @@ struct virtio_mmio_softc {
 
 struct virtio_mmio_attach_args {
 	struct virtio_attach_args	 vma_va;
-	struct fdt_attach_args          *vma_fa;
+	struct fdt_attach_args		*vma_fa;
 };
 
 const struct cfattach virtio_mmio_ca = {
@@ -160,6 +162,7 @@ const struct virtio_ops virtio_mmio_ops = {
 	virtio_mmio_attach_finish,
 	virtio_mmio_intr,
 	virtio_mmio_intr_barrier,
+	virtio_mmio_intr_establish,
 };
 
 uint16_t
@@ -234,7 +237,7 @@ virtio_mmio_set_status(struct virtio_softc *vsc, int status)
 		    VIRTIO_MMIO_STATUS) != 0) {
 			CPU_BUSY_CYCLE();
 		}
-	} else  {
+	} else {
 		old = bus_space_read_4(sc->sc_iot, sc->sc_ioh,
 		    VIRTIO_MMIO_STATUS);
 		bus_space_write_4(sc->sc_iot, sc->sc_ioh, VIRTIO_MMIO_STATUS,
@@ -546,3 +549,11 @@ virtio_mmio_intr_barrier(struct virtio_softc *vsc)
 	if (sc->sc_ih)
 		intr_barrier(sc->sc_ih);
 }
+
+int
+virtio_mmio_intr_establish(struct virtio_softc *vsc,
+    struct virtio_attach_args *va, int vec, struct cpu_info *ci,
+    int (*func)(void *), void *arg)
+{
+	return ENXIO;
+}

sys/dev/pci/drm/i915/gt/intel_context.c

@@ -404,7 +404,7 @@ intel_context_init(struct intel_context *ce, struct intel_engine_cs *engine)
 
 	rw_init(&ce->pin_mutex, "cepin");
 
-	mtx_init(&ce->guc_state.lock, IPL_NONE);
+	mtx_init(&ce->guc_state.lock, IPL_TTY);
 	INIT_LIST_HEAD(&ce->guc_state.fences);
 	INIT_LIST_HEAD(&ce->guc_state.requests);
 

sys/dev/pci/drm/i915/gt/intel_engine_user.c

@@ -165,7 +165,7 @@ static void set_scheduler_caps(struct drm_i915_private *i915)
 			disabled |= (I915_SCHEDULER_CAP_ENABLED |
 				     I915_SCHEDULER_CAP_PRIORITY);
 
-		if (intel_uc_uses_guc_submission(&to_gt(i915)->uc))
+		if (intel_uc_uses_guc_submission(&engine->gt->uc))
 			enabled |= I915_SCHEDULER_CAP_STATIC_PRIORITY_MAP;
 
 		for (i = 0; i < ARRAY_SIZE(map); i++) {

sys/dev/pci/virtio_pci.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: virtio_pci.c,v 1.48 2024/12/20 22:18:27 sf Exp $	*/
+/*	$OpenBSD: virtio_pci.c,v 1.50 2025/01/14 14:28:38 sf Exp $	*/
 /*	$NetBSD: virtio.c,v 1.3 2011/11/02 23:05:52 njoly Exp $	*/
 
 /*
@@ -50,7 +50,7 @@
  * XXX: PCI-endian while the device specific registers are native endian.
  */
 
-#define MAX_MSIX_VECS	8
+#define MAX_MSIX_VECS	16
 
 struct virtio_pci_softc;
 struct virtio_pci_attach_args;
@@ -62,7 +62,7 @@ int		virtio_pci_attach_10(struct virtio_pci_softc *sc, struct pci_attach_args *p
 int		virtio_pci_detach(struct device *, int);
 
 void		virtio_pci_kick(struct virtio_softc *, uint16_t);
-int		virtio_pci_adjust_config_region(struct virtio_pci_softc *);
+int		virtio_pci_adjust_config_region(struct virtio_pci_softc *, int offset);
 uint8_t		virtio_pci_read_device_config_1(struct virtio_softc *, int);
 uint16_t	virtio_pci_read_device_config_2(struct virtio_softc *, int);
 uint32_t	virtio_pci_read_device_config_4(struct virtio_softc *, int);
@@ -81,9 +81,10 @@ int		virtio_pci_negotiate_features(struct virtio_softc *, const struct virtio_fe
 int		virtio_pci_negotiate_features_10(struct virtio_softc *, const struct virtio_feature_name *);
 void		virtio_pci_set_msix_queue_vector(struct virtio_pci_softc *, uint32_t, uint16_t);
 void		virtio_pci_set_msix_config_vector(struct virtio_pci_softc *, uint16_t);
-int		virtio_pci_msix_establish(struct virtio_pci_softc *, struct virtio_pci_attach_args *, int, int (*)(void *), void *);
+int		virtio_pci_msix_establish(struct virtio_pci_softc *, struct virtio_pci_attach_args *, int, struct cpu_info *, int (*)(void *), void *);
 int		virtio_pci_setup_msix(struct virtio_pci_softc *, struct virtio_pci_attach_args *, int);
 void		virtio_pci_intr_barrier(struct virtio_softc *);
+int		virtio_pci_intr_establish(struct virtio_softc *, struct virtio_attach_args *, int, struct cpu_info *, int (*)(void *), void *);
 void		virtio_pci_free_irqs(struct virtio_pci_softc *);
 int		virtio_pci_poll_intr(void *);
 int		virtio_pci_legacy_intr(void *);
@@ -100,6 +101,7 @@ enum irq_type {
 	IRQ_NO_MSIX,
 	IRQ_MSIX_SHARED, /* vec 0: config irq, vec 1 shared by all vqs */
 	IRQ_MSIX_PER_VQ, /* vec 0: config irq, vec n: irq of vq[n-1] */
+	IRQ_MSIX_CHILD,  /* assigned by child driver */
 };
 
 struct virtio_pci_intr {
@@ -179,6 +181,7 @@ const struct virtio_ops virtio_pci_ops = {
 	virtio_pci_attach_finish,
 	virtio_pci_poll_intr,
 	virtio_pci_intr_barrier,
+	virtio_pci_intr_establish,
 };
 
 static inline uint64_t
@@ -426,7 +429,7 @@ virtio_pci_find_cap(struct virtio_pci_softc *sc, int cfg_type, void *buf, int bu
 			printf("%s: cap too large\n", __func__);
 			return ERANGE;
 		}
-		for (i = 4; i < len / sizeof(pcireg_t);  i++)
+		for (i = 4; i < len / sizeof(pcireg_t); i++)
 			v->reg[i] = pci_conf_read(pc, tag, offset + i * 4);
 	}
 
@@ -648,10 +651,12 @@ virtio_pci_attach(struct device *parent, struct device *self, void *aux)
 		goto free;
 	}
 
-	sc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
 	sc->sc_irq_type = IRQ_NO_MSIX;
-	if (virtio_pci_adjust_config_region(sc) != 0)
-		goto err;
+	if (virtio_pci_adjust_config_region(sc,
+	    VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI) != 0)
+	{
+		goto free;
+	}
 
 	virtio_device_reset(vsc);
 	virtio_set_status(vsc, VIRTIO_CONFIG_DEVICE_STATUS_ACK);
@@ -692,7 +697,9 @@ virtio_pci_attach_finish(struct virtio_softc *vsc,
 	pci_chipset_tag_t pc = vpa->vpa_pa->pa_pc;
 	char const *intrstr;
 
-	if (virtio_pci_setup_msix(sc, vpa, 0) == 0) {
+	if (sc->sc_irq_type == IRQ_MSIX_CHILD) {
+		intrstr = "msix";
+	} else if (virtio_pci_setup_msix(sc, vpa, 0) == 0) {
 		sc->sc_irq_type = IRQ_MSIX_PER_VQ;
 		intrstr = "msix per-VQ";
 	} else if (virtio_pci_setup_msix(sc, vpa, 1) == 0) {
@@ -754,11 +761,14 @@ virtio_pci_detach(struct device *self, int flags)
 }
 
 int
-virtio_pci_adjust_config_region(struct virtio_pci_softc *sc)
+virtio_pci_adjust_config_region(struct virtio_pci_softc *sc, int offset)
 {
 	if (sc->sc_sc.sc_version_1)
 		return 0;
-	sc->sc_devcfg_iosize = sc->sc_iosize - sc->sc_devcfg_offset;
+	if (sc->sc_devcfg_offset == offset)
+		return 0;
+	sc->sc_devcfg_offset = offset;
+	sc->sc_devcfg_iosize = sc->sc_iosize - offset;
 	sc->sc_devcfg_iot = sc->sc_iot;
 	if (bus_space_subregion(sc->sc_iot, sc->sc_ioh, sc->sc_devcfg_offset,
 	    sc->sc_devcfg_iosize, &sc->sc_devcfg_ioh) != 0) {
@@ -958,30 +968,33 @@ virtio_pci_write_device_config_8(struct virtio_softc *vsc,
 
 int
 virtio_pci_msix_establish(struct virtio_pci_softc *sc,
-    struct virtio_pci_attach_args *vpa, int idx,
+    struct virtio_pci_attach_args *vpa, int idx, struct cpu_info *ci,
     int (*handler)(void *), void *ih_arg)
 {
 	struct virtio_softc *vsc = &sc->sc_sc;
 	pci_intr_handle_t ih;
+	int r;
 
 	KASSERT(idx < sc->sc_nintr);
 
-	if (pci_intr_map_msix(vpa->vpa_pa, idx, &ih) != 0) {
+	r = pci_intr_map_msix(vpa->vpa_pa, idx, &ih);
+	if (r != 0) {
 #if VIRTIO_DEBUG
 		printf("%s[%d]: pci_intr_map_msix failed\n",
 		    vsc->sc_dev.dv_xname, idx);
 #endif
-		return 1;
+		return r;
 	}
 	snprintf(sc->sc_intr[idx].name, sizeof(sc->sc_intr[idx].name), "%s:%d",
 	    vsc->sc_child->dv_xname, idx);
-	sc->sc_intr[idx].ih = pci_intr_establish(sc->sc_pc, ih, vsc->sc_ipl,
-	    handler, ih_arg, sc->sc_intr[idx].name);
+	sc->sc_intr[idx].ih = pci_intr_establish_cpu(sc->sc_pc, ih, vsc->sc_ipl,
+	    ci, handler, ih_arg, sc->sc_intr[idx].name);
 	if (sc->sc_intr[idx].ih == NULL) {
 		printf("%s[%d]: couldn't establish msix interrupt\n",
-		    vsc->sc_dev.dv_xname, idx);
-		return 1;
+		    vsc->sc_child->dv_xname, idx);
+		return ENOMEM;
 	}
+	virtio_pci_adjust_config_region(sc, VIRTIO_CONFIG_DEVICE_CONFIG_MSI);
 	return 0;
 }
 
@@ -1031,8 +1044,8 @@ virtio_pci_free_irqs(struct virtio_pci_softc *sc)
 		}
 	}
 
-	sc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI;
-	virtio_pci_adjust_config_region(sc);
+	/* XXX msix_delroute does not unset PCI_MSIX_MC_MSIXE -> leave alone? */
+	virtio_pci_adjust_config_region(sc, VIRTIO_CONFIG_DEVICE_CONFIG_NOMSI);
 }
 
 int
@@ -1040,34 +1053,33 @@ virtio_pci_setup_msix(struct virtio_pci_softc *sc,
     struct virtio_pci_attach_args *vpa, int shared)
 {
 	struct virtio_softc *vsc = &sc->sc_sc;
-	int i;
+	int i, r = 0;
 
 	/* Shared needs config + queue */
 	if (shared && vpa->vpa_va.va_nintr < 1 + 1)
-		return 1;
+		return ERANGE;
 	/* Per VQ needs config + N * queue */
 	if (!shared && vpa->vpa_va.va_nintr < 1 + vsc->sc_nvqs)
-		return 1;
+		return ERANGE;
 
-	if (virtio_pci_msix_establish(sc, vpa, 0, virtio_pci_config_intr, vsc))
-		return 1;
-	sc->sc_devcfg_offset = VIRTIO_CONFIG_DEVICE_CONFIG_MSI;
-	virtio_pci_adjust_config_region(sc);
+	r = virtio_pci_msix_establish(sc, vpa, 0, NULL, virtio_pci_config_intr, vsc);
+	if (r != 0)
+		return r;
 
 	if (shared) {
-		if (virtio_pci_msix_establish(sc, vpa, 1,
-		    virtio_pci_shared_queue_intr, vsc)) {
+		r = virtio_pci_msix_establish(sc, vpa, 1, NULL,
+		    virtio_pci_shared_queue_intr, vsc);
+		if (r != 0)
 			goto fail;
-		}
 
 		for (i = 0; i < vsc->sc_nvqs; i++)
 			vsc->sc_vqs[i].vq_intr_vec = 1;
 	} else {
 		for (i = 0; i < vsc->sc_nvqs; i++) {
-			if (virtio_pci_msix_establish(sc, vpa, i + 1,
-			    virtio_pci_queue_intr, &vsc->sc_vqs[i])) {
+			r = virtio_pci_msix_establish(sc, vpa, i + 1, NULL,
+			    virtio_pci_queue_intr, &vsc->sc_vqs[i]);
+			if (r != 0)
 				goto fail;
-			}
 			vsc->sc_vqs[i].vq_intr_vec = i + 1;
 		}
 	}
@@ -1075,7 +1087,28 @@ virtio_pci_setup_msix(struct virtio_pci_softc *sc,
 	return 0;
 fail:
 	virtio_pci_free_irqs(sc);
-	return 1;
+	return r;
+}
+
+int
+virtio_pci_intr_establish(struct virtio_softc *vsc,
+    struct virtio_attach_args *va, int vec, struct cpu_info *ci,
+    int (*func)(void *), void *arg)
+{
+	struct virtio_pci_attach_args *vpa;
+	struct virtio_pci_softc *sc;
+
+	if (vsc->sc_ops != &virtio_pci_ops)
+		return ENXIO;
+
+	vpa = (struct virtio_pci_attach_args *)va;
+	sc = (struct virtio_pci_softc *)vsc;
+
+	if (vec >= sc->sc_nintr || sc->sc_nintr <= 1)
+		return ERANGE;
+
+	sc->sc_irq_type = IRQ_MSIX_CHILD;
+	return virtio_pci_msix_establish(sc, vpa, vec, ci, func, arg);
 }
 
 void

sys/dev/pv/if_vio.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_vio.c,v 1.63 2025/01/06 14:23:52 sf Exp $	*/
+/*	$OpenBSD: if_vio.c,v 1.66 2025/01/14 14:32:32 sf Exp $	*/
 
 /*
  * Copyright (c) 2012 Stefan Fritsch, Alexander Fiveg.
@@ -32,8 +32,10 @@
 #include <sys/param.h>
 #include <sys/systm.h>
 #include <sys/device.h>
+#include <sys/intrmap.h>
 #include <sys/mbuf.h>
 #include <sys/mutex.h>
+#include <sys/percpu.h>	/* for CACHELINESIZE */
 #include <sys/sockio.h>
 #include <sys/timeout.h>
 
@@ -64,14 +66,21 @@
  * if_vioreg.h:
  */
 /* Configuration registers */
-#define VIRTIO_NET_CONFIG_MAC		0 /* 8bit x 6byte */
-#define VIRTIO_NET_CONFIG_STATUS	6 /* 16bit */
+#define VIRTIO_NET_CONFIG_MAC		 0 /*  8 bit x 6 byte */
+#define VIRTIO_NET_CONFIG_STATUS	 6 /* 16 bit */
+#define VIRTIO_NET_CONFIG_MAX_QUEUES	 8 /* 16 bit */
+#define VIRTIO_NET_CONFIG_MTU		10 /* 16 bit */
+#define VIRTIO_NET_CONFIG_SPEED		12 /* 32 bit */
+#define VIRTIO_NET_CONFIG_DUPLEX	16 /*  8 bit */
+#define VIRTIO_NET_CONFIG_RSS_SIZE	17 /*  8 bit */
+#define VIRTIO_NET_CONFIG_RSS_LEN	18 /* 16 bit */
+#define VIRTIO_NET_CONFIG_HASH_TYPES	20 /* 16 bit */
 
 /* Feature bits */
 #define VIRTIO_NET_F_CSUM			(1ULL<<0)
 #define VIRTIO_NET_F_GUEST_CSUM			(1ULL<<1)
-#define VIRTIO_NET_F_CTRL_GUEST_OFFLOADS        (1ULL<<2)
-#define VIRTIO_NET_F_MTU                        (1ULL<<3)
+#define VIRTIO_NET_F_CTRL_GUEST_OFFLOADS	(1ULL<<2)
+#define VIRTIO_NET_F_MTU			(1ULL<<3)
 #define VIRTIO_NET_F_MAC			(1ULL<<5)
 #define VIRTIO_NET_F_GSO			(1ULL<<6)
 #define VIRTIO_NET_F_GUEST_TSO4			(1ULL<<7)
@@ -183,6 +192,11 @@ struct virtio_net_ctrl_cmd {
 # define VIRTIO_NET_CTRL_VLAN_ADD	0
 # define VIRTIO_NET_CTRL_VLAN_DEL	1
 
+#define VIRTIO_NET_CTRL_MQ		4
+# define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET	0
+# define VIRTIO_NET_CTRL_MQ_RSS_CONFIG		1
+# define VIRTIO_NET_CTRL_MQ_HASH_CONFIG		2
+
 #define VIRTIO_NET_CTRL_GUEST_OFFLOADS	5
 # define VIRTIO_NET_CTRL_GUEST_OFFLOADS_SET	0
 
@@ -196,6 +210,12 @@ struct virtio_net_ctrl_rx {
 	uint8_t	onoff;
 } __packed;
 
+struct virtio_net_ctrl_mq_pairs_set {
+	uint16_t virtqueue_pairs;
+};
+#define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MIN	1
+#define VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX	0x8000
+
 struct virtio_net_ctrl_guest_offloads {
 	uint64_t offloads;
 } __packed;
@@ -231,7 +251,7 @@ struct vio_queue {
 	struct virtqueue	 *viq_txvq;
 	struct mutex		  viq_txmtx, viq_rxmtx;
 	int			  viq_txfree_slots;
-};
+} __aligned(CACHELINESIZE);
 
 struct vio_softc {
 	struct device		sc_dev;
@@ -251,14 +271,16 @@ struct vio_softc {
 	caddr_t			sc_dma_kva;
 
 	int			sc_hdr_size;
-	struct virtio_net_ctrl_cmd *sc_ctrl_cmd;
-	struct virtio_net_ctrl_status *sc_ctrl_status;
-	struct virtio_net_ctrl_rx *sc_ctrl_rx;
-	struct virtio_net_ctrl_guest_offloads *sc_ctrl_guest_offloads;
-	struct virtio_net_ctrl_mac_tbl *sc_ctrl_mac_tbl_uc;
+	struct virtio_net_ctrl_cmd		*sc_ctrl_cmd;
+	struct virtio_net_ctrl_status		*sc_ctrl_status;
+	struct virtio_net_ctrl_rx		*sc_ctrl_rx;
+	struct virtio_net_ctrl_mq_pairs_set	*sc_ctrl_mq_pairs;
+	struct virtio_net_ctrl_guest_offloads	*sc_ctrl_guest_offloads;
+	struct virtio_net_ctrl_mac_tbl		*sc_ctrl_mac_tbl_uc;
 #define sc_ctrl_mac_info sc_ctrl_mac_tbl_uc
-	struct virtio_net_ctrl_mac_tbl *sc_ctrl_mac_tbl_mc;
+	struct virtio_net_ctrl_mac_tbl		*sc_ctrl_mac_tbl_mc;
 
+	struct intrmap		*sc_intrmap;
 	struct vio_queue	*sc_q;
 	uint16_t		sc_nqueues;
 	int			sc_tx_slots_per_req;
@@ -317,10 +339,15 @@ void	vio_tx_drain(struct vio_softc *);
 int	vio_encap(struct vio_queue *, int, struct mbuf *);
 void	vio_txtick(void *);
 
+int	vio_queue_intr(void *);
+int	vio_config_intr(void *);
+int	vio_ctrl_intr(void *);
+
 /* other control */
 void	vio_link_state(struct ifnet *);
 int	vio_config_change(struct virtio_softc *);
 int	vio_ctrl_rx(struct vio_softc *, int, int);
+int	vio_ctrl_mq(struct vio_softc *);
 int	vio_ctrl_guest_offloads(struct vio_softc *, uint64_t);
 int	vio_set_rx_filter(struct vio_softc *);
 void	vio_iff(struct vio_softc *);
@@ -408,6 +435,8 @@ vio_free_dmamem(struct vio_softc *sc)
  *   sc_ctrl_status:	 return value for a command via ctrl vq (READ)
  *   sc_ctrl_rx:	 parameter for a VIRTIO_NET_CTRL_RX class command
  *			 (WRITE)
+ *   sc_ctrl_mq_pairs_set: set number of rx/tx queue pais (WRITE)
+ *   sc_ctrl_guest_offloads: configure offload features (WRITE)
  *   sc_ctrl_mac_tbl_uc: unicast MAC address filter for a VIRTIO_NET_CTRL_MAC
  *			 class command (WRITE)
  *   sc_ctrl_mac_tbl_mc: multicast MAC address filter for a VIRTIO_NET_CTRL_MAC
@@ -449,6 +478,7 @@ vio_alloc_mem(struct vio_softc *sc, int tx_max_segments)
 		allocsize += sizeof(struct virtio_net_ctrl_cmd) * 1;
 		allocsize += sizeof(struct virtio_net_ctrl_status) * 1;
 		allocsize += sizeof(struct virtio_net_ctrl_rx) * 1;
+		allocsize += sizeof(struct virtio_net_ctrl_mq_pairs_set) * 1;
 		allocsize += sizeof(struct virtio_net_ctrl_guest_offloads) * 1;
 		allocsize += VIO_CTRL_MAC_INFO_SIZE;
 	}
@@ -474,6 +504,8 @@ vio_alloc_mem(struct vio_softc *sc, int tx_max_segments)
 		offset += sizeof(*sc->sc_ctrl_status);
 		sc->sc_ctrl_rx = (void *)(kva + offset);
 		offset += sizeof(*sc->sc_ctrl_rx);
+		sc->sc_ctrl_mq_pairs = (void *)(kva + offset);
+		offset += sizeof(*sc->sc_ctrl_mq_pairs);
 		sc->sc_ctrl_guest_offloads = (void *)(kva + offset);
 		offset += sizeof(*sc->sc_ctrl_guest_offloads);
 		sc->sc_ctrl_mac_tbl_uc = (void *)(kva + offset);
@@ -598,7 +630,7 @@ vio_attach(struct device *parent, struct device *self, void *aux)
 	struct vio_softc *sc = (struct vio_softc *)self;
 	struct virtio_softc *vsc = (struct virtio_softc *)parent;
 	struct virtio_attach_args *va = aux;
-	int i, tx_max_segments;
+	int i, r, tx_max_segments;
 	struct ifnet *ifp = &sc->sc_ac.ac_if;
 
 	if (vsc->sc_child != NULL) {
@@ -616,6 +648,9 @@ vio_attach(struct device *parent, struct device *self, void *aux)
 	    VIRTIO_NET_F_MRG_RXBUF | VIRTIO_NET_F_CSUM |
 	    VIRTIO_F_RING_EVENT_IDX | VIRTIO_NET_F_GUEST_CSUM;
 
+	if (va->va_nintr > 3 && ncpus > 1)
+		vsc->sc_driver_features |= VIRTIO_NET_F_MQ;
+
 	vsc->sc_driver_features |= VIRTIO_NET_F_HOST_TSO4;
 	vsc->sc_driver_features |= VIRTIO_NET_F_HOST_TSO6;
 
@@ -626,10 +661,23 @@ vio_attach(struct device *parent, struct device *self, void *aux)
 	if (virtio_negotiate_features(vsc, virtio_net_feature_names) != 0)
 		goto err;
 
-	sc->sc_nqueues = 1;
-	vsc->sc_nvqs = 2 * sc->sc_nqueues;
-	if (virtio_has_feature(vsc, VIRTIO_NET_F_CTRL_VQ))
-		vsc->sc_nvqs++;
+	if (virtio_has_feature(vsc, VIRTIO_NET_F_MQ)) {
+		i = virtio_read_device_config_2(vsc,
+		    VIRTIO_NET_CONFIG_MAX_QUEUES);
+		vsc->sc_nvqs = 2 * i + 1;
+		i = MIN(i, VIRTIO_NET_CTRL_MQ_VQ_PAIRS_MAX);
+		sc->sc_intrmap = intrmap_create(&sc->sc_dev, i,
+		    va->va_nintr - 2, 0);
+		sc->sc_nqueues = intrmap_count(sc->sc_intrmap);
+		printf(": %u queue%s", sc->sc_nqueues,
+		    sc->sc_nqueues > 1 ? "s" : "");
+	} else {
+		sc->sc_nqueues = 1;
+		printf(": 1 queue");
+		vsc->sc_nvqs = 2;
+		if (virtio_has_feature(vsc, VIRTIO_NET_F_CTRL_VQ))
+			vsc->sc_nvqs++;
+	}
 
 	vsc->sc_vqs = mallocarray(vsc->sc_nvqs, sizeof(*vsc->sc_vqs), M_DEVBUF,
 	    M_WAITOK|M_ZERO);
@@ -729,18 +777,66 @@ vio_attach(struct device *parent, struct device *self, void *aux)
 		else
 			virtio_stop_vq_intr(vsc, vioq->viq_txvq);
 		vioq->viq_txfree_slots = vioq->viq_txvq->vq_num - 1;
+		KASSERT(vioq->viq_txfree_slots > sc->sc_tx_slots_per_req);
+		if (vioq->viq_txvq->vq_num != sc->sc_q[0].viq_txvq->vq_num) {
+			printf("inequal tx queue size %d: %d != %d\n", i,
+			    vioq->viq_txvq->vq_num,
+			    sc->sc_q[0].viq_txvq->vq_num);
+			goto err;
+		}
+		DPRINTF("%d: q %p rx %p tx %p\n", i, vioq, vioq->viq_rxvq,
+		    vioq->viq_txvq);
+
+		if (sc->sc_intrmap != NULL) {
+			vioq->viq_rxvq->vq_intr_vec = i + 2;
+			vioq->viq_txvq->vq_intr_vec = i + 2;
+		}
 	}
 
 	/* control queue */
 	if (virtio_has_feature(vsc, VIRTIO_NET_F_CTRL_VQ)) {
-		sc->sc_ctl_vq = &vsc->sc_vqs[2];
-		if (virtio_alloc_vq(vsc, sc->sc_ctl_vq, 2, 1,
-		    "control") != 0)
+		i = 2;
+		if (virtio_has_feature(vsc, VIRTIO_NET_F_MQ)) {
+			i = 2 * virtio_read_device_config_2(vsc,
+			    VIRTIO_NET_CONFIG_MAX_QUEUES);
+		}
+		sc->sc_ctl_vq = &vsc->sc_vqs[i];
+		if (virtio_alloc_vq(vsc, sc->sc_ctl_vq, i, 1, "control") != 0)
 			goto err;
 		sc->sc_ctl_vq->vq_done = vio_ctrleof;
+		if (sc->sc_intrmap != NULL)
+			sc->sc_ctl_vq->vq_intr_vec = 1;
 		virtio_start_vq_intr(vsc, sc->sc_ctl_vq);
 	}
 
+	if (sc->sc_intrmap) {
+		r = virtio_intr_establish(vsc, va, 0, NULL, vio_config_intr,
+		    vsc);
+		if (r != 0) {
+			printf("%s: cannot alloc config intr: %d\n",
+			    sc->sc_dev.dv_xname, r);
+			goto err;
+		}
+		r = virtio_intr_establish(vsc, va, 1, NULL, vio_ctrl_intr,
+		    sc->sc_ctl_vq);
+		if (r != 0) {
+			printf("%s: cannot alloc ctrl intr: %d\n",
+			    sc->sc_dev.dv_xname, r);
+			goto err;
+		}
+		for (i = 0; i < sc->sc_nqueues; i++) {
+			struct cpu_info *ci = NULL;
+			ci = intrmap_cpu(sc->sc_intrmap, i);
+			r = virtio_intr_establish(vsc, va, i + 2, ci,
+			    vio_queue_intr, &sc->sc_q[i]);
+			if (r != 0) {
+				printf("%s: cannot alloc q%d intr: %d\n",
+				    sc->sc_dev.dv_xname, i, r);
+				goto err;
+			}
+		}
+	}
+
 	if (vio_alloc_mem(sc, tx_max_segments) < 0)
 		goto err;
 
@@ -760,6 +856,11 @@ vio_attach(struct device *parent, struct device *self, void *aux)
 	if (virtio_attach_finish(vsc, va) != 0)
 		goto err;
 
+	if (virtio_has_feature(vsc, VIRTIO_NET_F_MQ)) {
+		/* ctrl queue works only after DRIVER_OK */
+		vio_ctrl_mq(sc);
+	}
+
 	if_attach(ifp);
 	ether_ifattach(ifp);
 	vio_link_state(ifp);
@@ -805,6 +906,33 @@ vio_link_state(struct ifnet *ifp)
 	}
 }
 
+/* interrupt handlers for multi-queue */
+int
+vio_queue_intr(void *arg)
+{
+	struct vio_queue *vioq = arg;
+	struct virtio_softc *vsc = vioq->viq_sc->sc_virtio;
+	int r;
+	r = virtio_check_vq(vsc, vioq->viq_txvq);
+	r |= virtio_check_vq(vsc, vioq->viq_rxvq);
+	return r;
+}
+
+int
+vio_config_intr(void *arg)
+{
+	struct virtio_softc *vsc = arg;
+	return vio_config_change(vsc);
+}
+
+int
+vio_ctrl_intr(void *arg)
+{
+	struct virtqueue *vq = arg;
+	return virtio_check_vq(vq->vq_owner, vq);
+}
+
+
 int
 vio_config_change(struct virtio_softc *vsc)
 {
@@ -913,6 +1041,8 @@ vio_stop(struct ifnet *ifp, int disable)
 	if (virtio_has_feature(vsc, VIRTIO_NET_F_CTRL_VQ))
 		virtio_start_vq_intr(vsc, sc->sc_ctl_vq);
 	virtio_reinit_end(vsc);
+	if (virtio_has_feature(vsc, VIRTIO_NET_F_MQ))
+		vio_ctrl_mq(sc);
 	if (virtio_has_feature(vsc, VIRTIO_NET_F_CTRL_VQ))
 		vio_ctrl_wakeup(sc, FREE);
 }
@@ -1137,6 +1267,33 @@ vio_dump(struct vio_softc *sc)
 }
 #endif
 
+static int
+vio_rxr_info(struct vio_softc *sc, struct if_rxrinfo *ifri)
+{
+	struct if_rxring_info *ifrs, *ifr;
+	int error;
+	unsigned int i;
+
+	ifrs = mallocarray(sc->sc_nqueues, sizeof(*ifrs),
+	    M_TEMP, M_WAITOK|M_ZERO|M_CANFAIL);
+	if (ifrs == NULL)
+		return (ENOMEM);
+
+	for (i = 0; i < sc->sc_nqueues; i++) {
+		ifr = &ifrs[i];
+
+		ifr->ifr_size = sc->sc_rx_mbuf_size;
+		snprintf(ifr->ifr_name, sizeof(ifr->ifr_name), "%u", i);
+		ifr->ifr_info = sc->sc_q[i].viq_rxring;
+	}
+
+	error = if_rxr_info_ioctl(ifri, i, ifrs);
+
+	free(ifrs, M_TEMP, i * sizeof(*ifrs));
+
+	return (error);
+}
+
 int
 vio_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 {
@@ -1171,8 +1328,7 @@ vio_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 		r = ifmedia_ioctl(ifp, ifr, &sc->sc_media, cmd);
 		break;
 	case SIOCGIFRXR:
-		r = if_rxr_ioctl((struct if_rxrinfo *)ifr->ifr_data,
-		    NULL, sc->sc_rx_mbuf_size, &sc->sc_q[0].viq_rxring);
+		r = vio_rxr_info(sc, (struct if_rxrinfo *)ifr->ifr_data);
 		break;
 	default:
 		r = ether_ioctl(ifp, &sc->sc_ac, cmd, data);
@@ -1666,6 +1822,8 @@ vio_ctrl_submit(struct vio_softc *sc, int slot)
 			vio_ctrl_wakeup(sc, RESET);
 			return ENXIO;
 		}
+		if (cold)
+			virtio_check_vq(sc->sc_virtio, sc->sc_ctl_vq);
 	}
 
 	VIO_DMAMEM_SYNC(vsc, sc, sc->sc_ctrl_cmd,
@@ -1723,6 +1881,41 @@ vio_ctrl_rx(struct vio_softc *sc, int cmd, int onoff)
 	return r;
 }
 
+/* issue a VIRTIO_NET_CTRL_MQ class command and wait for completion */
+int
+vio_ctrl_mq(struct vio_softc *sc)
+{
+	struct virtio_softc *vsc = sc->sc_virtio;
+	struct virtqueue *vq = sc->sc_ctl_vq;
+	int r, slot;
+
+
+	r = vio_ctrl_start(sc, VIRTIO_NET_CTRL_MQ,
+	    VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, 1, &slot);
+	if (r != 0)
+		return r;
+
+	sc->sc_ctrl_mq_pairs->virtqueue_pairs = sc->sc_nqueues;
+
+	vio_dmamem_enqueue(vsc, sc, vq, slot, sc->sc_ctrl_mq_pairs,
+	    sizeof(*sc->sc_ctrl_mq_pairs), 1);
+
+	r = vio_ctrl_submit(sc, slot);
+
+	VIO_DMAMEM_SYNC(vsc, sc, sc->sc_ctrl_mq_pairs,
+	    sizeof(*sc->sc_ctrl_mq_pairs), BUS_DMASYNC_POSTWRITE);
+
+	if (r != 0)
+		printf("%s: ctrl cmd %d failed\n", sc->sc_dev.dv_xname,
+		    VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET);
+
+	DPRINTF("%s: cmd %d %d: %d\n", __func__,
+	    VIRTIO_NET_CTRL_MQ_VQ_PAIRS_SET, sc->sc_nqueues, r);
+
+	vio_ctrl_finish(sc);
+	return r;
+}
+
 int
 vio_ctrl_guest_offloads(struct vio_softc *sc, uint64_t features)
 {
@@ -1745,7 +1938,7 @@ vio_ctrl_guest_offloads(struct vio_softc *sc, uint64_t features)
 	VIO_DMAMEM_SYNC(vsc, sc, sc->sc_ctrl_guest_offloads,
 	    sizeof(*sc->sc_ctrl_guest_offloads), BUS_DMASYNC_POSTWRITE);
 
-	if (r != 0) {
+	if (r != 0 && features != 0) {
 		printf("%s: offload features 0x%llx failed\n",
 		    sc->sc_dev.dv_xname, features);
 	}

sys/dev/pv/virtiovar.h

@@ -1,4 +1,4 @@
-/*	$OpenBSD: virtiovar.h,v 1.27 2025/01/09 10:55:22 sf Exp $	*/
+/*	$OpenBSD: virtiovar.h,v 1.28 2025/01/14 12:30:57 sf Exp $	*/
 /*	$NetBSD: virtiovar.h,v 1.1 2011/10/30 12:12:21 hannken Exp $	*/
 
 /*
@@ -165,6 +165,8 @@ struct virtio_ops {
 	int		(*attach_finish)(struct virtio_softc *, struct virtio_attach_args *);
 	int		(*poll_intr)(void *);
 	void		(*intr_barrier)(struct virtio_softc *);
+	int		(*intr_establish)(struct virtio_softc *, struct virtio_attach_args *,
+			    int, struct cpu_info *, int (*)(void *), void *);
 };
 
 #define VIRTIO_CHILD_ERROR	((void*)1)
@@ -208,6 +210,14 @@ struct virtio_softc {
 #define	virtio_set_status(sc, i)		(sc)->sc_ops->set_status(sc, i)
 #define	virtio_intr_barrier(sc)			(sc)->sc_ops->intr_barrier(sc)
 
+/*
+ * virtio_intr_establish() only works if va_nintr > 1. If it is called by a
+ * child driver, the transport driver will skip automatic intr allocation and
+ * the child driver must allocate all required interrupts itself. Vector 0 is
+ * always used for the config change interrupt.
+ */
+#define	virtio_intr_establish(sc, va, v, ci, fn, a)	(sc)->sc_ops->intr_establish(sc, va, v, ci, fn, a)
+
 /* only for transport drivers */
 #define	virtio_device_reset(sc)			virtio_set_status((sc), 0)
 

sys/kern/kern_malloc.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_malloc.c,v 1.152 2024/06/26 01:40:49 jsg Exp $	*/
+/*	$OpenBSD: kern_malloc.c,v 1.153 2025/01/14 18:37:51 mvs Exp $	*/
 /*	$NetBSD: kern_malloc.c,v 1.15.4.2 1996/06/13 17:10:56 cgd Exp $	*/
 
 /*
@@ -50,6 +50,11 @@
 #include <ddb/db_output.h>
 #endif
 
+/*
+ * Locks used to protect data:
+ *	I	Immutable data
+ */
+ 
 static
 #ifndef SMALL_KERNEL
 __inline__
@@ -95,12 +100,11 @@ struct kmemstats kmemstats[M_LAST];
 #endif
 struct kmemusage *kmemusage;
 char *kmembase, *kmemlimit;
-char buckstring[16 * sizeof("123456,")];
+char buckstring[16 * sizeof("123456,")];	/* [I] */
 int buckstring_init = 0;
 #if defined(KMEMSTATS) || defined(DIAGNOSTIC)
 char *memname[] = INITKMEMNAMES;
-char *memall = NULL;
-struct rwlock sysctl_kmemlock = RWLOCK_INITIALIZER("sysctlklk");
+char *memall;					/* [I] */
 #endif
 
 /*
@@ -540,6 +544,10 @@ kmeminit(void)
 	vaddr_t base, limit;
 	long indx;
 
+#if defined(KMEMSTATS) || defined(DIAGNOSTIC)
+	int i, siz, totlen;
+#endif
+
 #ifdef DIAGNOSTIC
 	if (sizeof(struct kmem_freelist) > (1 << MINBUCKET))
 		panic("kmeminit: minbucket too small/struct freelist too big");
@@ -577,6 +585,38 @@ kmeminit(void)
 	for (indx = 0; indx < M_LAST; indx++)
 		kmemstats[indx].ks_limit =
 		    (long)nkmempages * PAGE_SIZE * 6 / 10;
+
+	memset(buckstring, 0, sizeof(buckstring));
+	for (siz = 0, i = MINBUCKET; i < MINBUCKET + 16; i++) {
+		snprintf(buckstring + siz, sizeof buckstring - siz,
+		    "%d,", (u_int)(1<<i));
+		siz += strlen(buckstring + siz);
+	}
+	/* Remove trailing comma */
+	if (siz)
+		buckstring[siz - 1] = '\0';
+#endif
+#if defined(KMEMSTATS) || defined(DIAGNOSTIC)
+	/* Figure out how large a buffer we need */
+	for (totlen = 0, i = 0; i < M_LAST; i++) {
+		if (memname[i])
+			totlen += strlen(memname[i]);
+		totlen++;
+	}
+	memall = malloc(totlen + M_LAST, M_SYSCTL, M_WAITOK|M_ZERO);
+	for (siz = 0, i = 0; i < M_LAST; i++) {
+		snprintf(memall + siz, totlen + M_LAST - siz, "%s,",
+		    memname[i] ? memname[i] : "");
+		siz += strlen(memall + siz);
+	}
+	/* Remove trailing comma */
+	if (siz)
+		memall[siz - 1] = '\0';
+	/* Now, convert all spaces to underscores */
+	for (i = 0; i < totlen; i++) {
+		if (memall[i] == ' ')
+			memall[i] = '_';
+	}
 #endif
 }
 
@@ -591,10 +631,6 @@ sysctl_malloc(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
 #ifdef KMEMSTATS
 	struct kmemstats km;
 #endif
-#if defined(KMEMSTATS) || defined(DIAGNOSTIC)
-	int error;
-#endif
-	int i, siz;
 
 	if (namelen != 2 && name[0] != KERN_MALLOC_BUCKETS &&
 	    name[0] != KERN_MALLOC_KMEMNAMES)
@@ -602,20 +638,6 @@ sysctl_malloc(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
 
 	switch (name[0]) {
 	case KERN_MALLOC_BUCKETS:
-		/* Initialize the first time */
-		if (buckstring_init == 0) {
-			buckstring_init = 1;
-			memset(buckstring, 0, sizeof(buckstring));
-			for (siz = 0, i = MINBUCKET; i < MINBUCKET + 16; i++) {
-				snprintf(buckstring + siz,
-				    sizeof buckstring - siz,
-				    "%d,", (u_int)(1<<i));
-				siz += strlen(buckstring + siz);
-			}
-			/* Remove trailing comma */
-			if (siz)
-				buckstring[siz - 1] = '\0';
-		}
 		return (sysctl_rdstring(oldp, oldlenp, newp, buckstring));
 
 	case KERN_MALLOC_BUCKET:
@@ -635,41 +657,9 @@ sysctl_malloc(int *name, u_int namelen, void *oldp, size_t *oldlenp, void *newp,
 #else
 		return (EOPNOTSUPP);
 #endif
-	case KERN_MALLOC_KMEMNAMES:
 #if defined(KMEMSTATS) || defined(DIAGNOSTIC)
-		error = rw_enter(&sysctl_kmemlock, RW_WRITE|RW_INTR);
-		if (error)
-			return (error);
-		if (memall == NULL) {
-			int totlen;
-
-			/* Figure out how large a buffer we need */
-			for (totlen = 0, i = 0; i < M_LAST; i++) {
-				if (memname[i])
-					totlen += strlen(memname[i]);
-				totlen++;
-			}
-			memall = malloc(totlen + M_LAST, M_SYSCTL,
-			    M_WAITOK|M_ZERO);
-			for (siz = 0, i = 0; i < M_LAST; i++) {
-				snprintf(memall + siz,
-				    totlen + M_LAST - siz,
-				    "%s,", memname[i] ? memname[i] : "");
-				siz += strlen(memall + siz);
-			}
-			/* Remove trailing comma */
-			if (siz)
-				memall[siz - 1] = '\0';
-
-			/* Now, convert all spaces to underscores */
-			for (i = 0; i < totlen; i++)
-				if (memall[i] == ' ')
-					memall[i] = '_';
-		}
-		rw_exit_write(&sysctl_kmemlock);
+	case KERN_MALLOC_KMEMNAMES:
 		return (sysctl_rdstring(oldp, oldlenp, newp, memall));
-#else
-		return (EOPNOTSUPP);
 #endif
 	default:
 		return (EOPNOTSUPP);

sys/kern/kern_sysctl.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: kern_sysctl.c,v 1.461 2025/01/13 18:09:24 mvs Exp $	*/
+/*	$OpenBSD: kern_sysctl.c,v 1.462 2025/01/14 18:37:51 mvs Exp $	*/
 /*	$NetBSD: kern_sysctl.c,v 1.17 1996/05/20 17:49:05 mrg Exp $	*/
 
 /*-
@@ -403,6 +403,9 @@ kern_sysctl_dirs(int top_name, int *name, u_int namelen,
 	int error;
 
 	switch (top_name) {
+	case KERN_MALLOCSTATS:
+		return (sysctl_malloc(name, namelen, oldp, oldlenp,
+		    newp, newlen, p));
 	case KERN_POOL:
 		return (sysctl_dopool(name, namelen, oldp, oldlenp));
 #if NAUDIO > 0
@@ -454,9 +457,6 @@ kern_sysctl_dirs_locked(int top_name, int *name, u_int namelen,
 		return (sysctl_doprof(name, namelen, oldp, oldlenp,
 		    newp, newlen));
 #endif
-	case KERN_MALLOCSTATS:
-		return (sysctl_malloc(name, namelen, oldp, oldlenp,
-		    newp, newlen, p));
 	case KERN_TTY:
 		return (sysctl_tty(name, namelen, oldp, oldlenp,
 		    newp, newlen));

sys/netinet/tcp_timer.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: tcp_timer.c,v 1.80 2025/01/05 12:18:48 bluhm Exp $	*/
+/*	$OpenBSD: tcp_timer.c,v 1.81 2025/01/14 13:49:44 bluhm Exp $	*/
 /*	$NetBSD: tcp_timer.c,v 1.14 1996/02/13 23:44:09 christos Exp $	*/
 
 /*
@@ -106,6 +106,36 @@ tcp_timer_init(void)
 		tcp_delack_msecs = TCP_DELACK_MSECS;
 }
 
+static inline int
+tcp_timer_enter(struct inpcb *inp, struct socket **so, struct tcpcb **tp,
+    u_int timer)
+{
+	KASSERT(timer < TCPT_NTIMERS);
+
+	NET_LOCK_SHARED();
+	*so = in_pcbsolock_ref(inp);
+	if (*so == NULL) {
+		*tp = NULL;
+		return -1;
+	}
+	*tp = intotcpcb(inp);
+	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
+	if (*tp == NULL || !ISSET((*tp)->t_flags, TF_TIMER << timer) ||
+	    timeout_pending(&(*tp)->t_timer[timer]))
+		return -1;
+	CLR((*tp)->t_flags, TF_TIMER << timer);
+
+	return 0;
+}
+
+static inline void
+tcp_timer_leave(struct inpcb *inp, struct socket *so)
+{
+	in_pcbsounlock_rele(inp, so);
+	NET_UNLOCK_SHARED();
+	in_pcbunref(inp);
+}
+
 /*
  * Callout to process delayed ACKs for a TCPCB.
  */
@@ -113,6 +143,7 @@ void
 tcp_timer_delack(void *arg)
 {
 	struct inpcb *inp = arg;
+	struct socket *so;
 	struct tcpcb *otp = NULL, *tp;
 	short ostate;
 
@@ -121,15 +152,10 @@ tcp_timer_delack(void *arg)
 	 * for whatever reason, it will restart the delayed
 	 * ACK callout.
 	 */
-	NET_LOCK();
-	tp = intotcpcb(inp);
-	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
-	if (tp == NULL || !ISSET(tp->t_flags, TF_TMR_DELACK) ||
-	    timeout_pending(&tp->t_timer[TCPT_DELACK]))
+	if (tcp_timer_enter(inp, &so, &tp, TCPT_DELACK))
 		goto out;
-	CLR(tp->t_flags, TF_TMR_DELACK);
 
-	if (inp->inp_socket->so_options & SO_DEBUG) {
+	if (so->so_options & SO_DEBUG) {
 		otp = tp;
 		ostate = tp->t_state;
 	}
@@ -138,8 +164,7 @@ tcp_timer_delack(void *arg)
 	if (otp)
 		tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_DELACK, 0);
  out:
-	NET_UNLOCK();
-	in_pcbunref(inp);
+	tcp_timer_leave(inp, so);
 }
 
 /*
@@ -199,19 +224,15 @@ void
 tcp_timer_rexmt(void *arg)
 {
 	struct inpcb *inp = arg;
+	struct socket *so;
 	struct tcpcb *otp = NULL, *tp;
-	uint32_t rto;
 	short ostate;
+	uint32_t rto;
 
-	NET_LOCK();
-	tp = intotcpcb(inp);
-	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
-	if (tp == NULL || !ISSET(tp->t_flags, TF_TMR_REXMT) ||
-	    timeout_pending(&tp->t_timer[TCPT_REXMT]))
+	if (tcp_timer_enter(inp, &so, &tp, TCPT_REXMT))
 		goto out;
-	CLR(tp->t_flags, TF_TMR_REXMT);
 
-	if ((tp->t_flags & TF_PMTUD_PEND) && inp &&
+	if ((tp->t_flags & TF_PMTUD_PEND) &&
 	    SEQ_GEQ(tp->t_pmtud_th_seq, tp->snd_una) &&
 	    SEQ_LT(tp->t_pmtud_th_seq, (int)(tp->snd_una + tp->t_maxseg))) {
 		struct sockaddr_in sin;
@@ -249,7 +270,7 @@ tcp_timer_rexmt(void *arg)
 		    tp->t_softerror : ETIMEDOUT);
 		goto out;
 	}
-	if (inp->inp_socket->so_options & SO_DEBUG) {
+	if (so->so_options & SO_DEBUG) {
 		otp = tp;
 		ostate = tp->t_state;
 	}
@@ -270,13 +291,13 @@ tcp_timer_rexmt(void *arg)
 	 * lots more sophisticated searching to find the right
 	 * value here...
 	 */
-	if (ip_mtudisc && inp &&
+	if (ip_mtudisc &&
 	    TCPS_HAVEESTABLISHED(tp->t_state) &&
 	    tp->t_rxtshift > TCP_MAXRXTSHIFT / 6) {
 		struct rtentry *rt = NULL;
 
 		/* No data to send means path mtu is not a problem */
-		if (!inp->inp_socket->so_snd.sb_cc)
+		if (!READ_ONCE(so->so_snd.sb_cc))
 			goto leave;
 
 		rt = in_pcbrtentry(inp);
@@ -391,31 +412,26 @@ tcp_timer_rexmt(void *arg)
 	if (otp)
 		tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_REXMT, 0);
  out:
-	NET_UNLOCK();
-	in_pcbunref(inp);
+	tcp_timer_leave(inp, so);
 }
 
 void
 tcp_timer_persist(void *arg)
 {
 	struct inpcb *inp = arg;
+	struct socket *so;
 	struct tcpcb *otp = NULL, *tp;
-	uint32_t rto;
 	short ostate;
 	uint64_t now;
+	uint32_t rto;
 
-	NET_LOCK();
-	tp = intotcpcb(inp);
-	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
-	if (tp == NULL || !ISSET(tp->t_flags, TF_TMR_PERSIST) ||
-	    timeout_pending(&tp->t_timer[TCPT_PERSIST]))
+	if (tcp_timer_enter(inp, &so, &tp, TCPT_PERSIST))
 		goto out;
-	CLR(tp->t_flags, TF_TMR_PERSIST);
 
 	if (TCP_TIMER_ISARMED(tp, TCPT_REXMT))
 		goto out;
 
-	if (inp->inp_socket->so_options & SO_DEBUG) {
+	if (so->so_options & SO_DEBUG) {
 		otp = tp;
 		ostate = tp->t_state;
 	}
@@ -445,26 +461,21 @@ tcp_timer_persist(void *arg)
 	if (otp)
 		tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_PERSIST, 0);
  out:
-	NET_UNLOCK();
-	in_pcbunref(inp);
+	tcp_timer_leave(inp, so);
 }
 
 void
 tcp_timer_keep(void *arg)
 {
 	struct inpcb *inp = arg;
+	struct socket *so;
 	struct tcpcb *otp = NULL, *tp;
 	short ostate;
 
-	NET_LOCK();
-	tp = intotcpcb(inp);
-	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
-	if (tp == NULL || !ISSET(tp->t_flags, TF_TMR_KEEP) ||
-	    timeout_pending(&tp->t_timer[TCPT_KEEP]))
+	if (tcp_timer_enter(inp, &so, &tp, TCPT_KEEP))
 		goto out;
-	CLR(tp->t_flags, TF_TMR_KEEP);
 
-	if (inp->inp_socket->so_options & SO_DEBUG) {
+	if (so->so_options & SO_DEBUG) {
 		otp = tp;
 		ostate = tp->t_state;
 	}
@@ -475,7 +486,7 @@ tcp_timer_keep(void *arg)
 		goto out;
 	}
 	if ((atomic_load_int(&tcp_always_keepalive) ||
-	    inp->inp_socket->so_options & SO_KEEPALIVE) &&
+	    so->so_options & SO_KEEPALIVE) &&
 	    tp->t_state <= TCPS_CLOSING) {
 		int maxidle;
 		uint64_t now;
@@ -509,28 +520,23 @@ tcp_timer_keep(void *arg)
 	if (otp)
 		tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_KEEP, 0);
  out:
-	NET_UNLOCK();
-	in_pcbunref(inp);
+	tcp_timer_leave(inp, so);
 }
 
 void
 tcp_timer_2msl(void *arg)
 {
 	struct inpcb *inp = arg;
+	struct socket *so;
 	struct tcpcb *otp = NULL, *tp;
 	short ostate;
-	int maxidle;
 	uint64_t now;
+	int maxidle;
 
-	NET_LOCK();
-	tp = intotcpcb(inp);
-	/* Ignore canceled timeouts or timeouts that have been rescheduled. */
-	if (tp == NULL || !ISSET(tp->t_flags, TF_TMR_2MSL) ||
-	    timeout_pending(&tp->t_timer[TCPT_2MSL]))
+	if (tcp_timer_enter(inp, &so, &tp, TCPT_2MSL))
 		goto out;
-	CLR(tp->t_flags, TF_TMR_2MSL);
 
-	if (inp->inp_socket->so_options & SO_DEBUG) {
+	if (so->so_options & SO_DEBUG) {
 		otp = tp;
 		ostate = tp->t_state;
 	}
@@ -546,8 +552,7 @@ tcp_timer_2msl(void *arg)
 	if (otp)
 		tcp_trace(TA_TIMER, ostate, tp, otp, NULL, TCPT_2MSL, 0);
  out:
-	NET_UNLOCK();
-	in_pcbunref(inp);
+	tcp_timer_leave(inp, so);
 }
 
 void

usr.sbin/bgpd/rde_rib.c

@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde_rib.c,v 1.266 2024/12/12 20:19:03 claudio Exp $ */
+/*	$OpenBSD: rde_rib.c,v 1.267 2025/01/14 12:24:23 claudio Exp $ */
 
 /*
  * Copyright (c) 2003, 2004 Claudio Jeker <claudio@openbsd.org>
@@ -1273,6 +1273,7 @@ prefix_adjout_update(struct prefix *p, struct rde_peer *peer,
 			/* nothing changed */
 			p->validation_state = state->vstate;
 			p->lastchange = getmonotime();
+			p->flags &= ~PREFIX_FLAG_STALE;
 			return;
 		}
 
@@ -1343,6 +1344,7 @@ prefix_adjout_withdraw(struct prefix *p)
 	/* already a withdraw, shortcut */
 	if (p->flags & PREFIX_FLAG_WITHDRAW) {
 		p->lastchange = getmonotime();
+		p->flags &= ~PREFIX_FLAG_STALE;
 		return;
 	}
 	/* pending update just got withdrawn */