2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (C) 2013 Emulex
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions are met:
10 * 1. Redistributions of source code must retain the above copyright notice,
11 * this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * 3. Neither the name of the Emulex Corporation nor the names of its
18 * contributors may be used to endorse or promote products derived from
19 * this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
22 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
25 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
26 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
27 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
28 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
29 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
30 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
31 * POSSIBILITY OF SUCH DAMAGE.
33 * Contact Information:
34 * freebsd-drivers@emulex.com
38 * Costa Mesa, CA 92626
43 #include "opt_inet6.h"
49 #define is_tso_pkt(m) (m->m_pkthdr.csum_flags & CSUM_TSO)
51 /* UE Status Low CSR */
52 static char *ue_status_low_desc[] = {
87 /* UE Status High CSR */
88 static char *ue_status_hi_desc[] = {
123 struct oce_common_cqe_info{
125 uint8_t l4_cksum_pass:1;
126 uint8_t ip_cksum_pass:1;
127 uint8_t ipv6_frame:1;
136 /* Driver entry points prototypes */
137 static int oce_probe(device_t dev);
138 static int oce_attach(device_t dev);
139 static int oce_detach(device_t dev);
140 static int oce_shutdown(device_t dev);
141 static int oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
142 static void oce_init(void *xsc);
143 static int oce_multiq_start(struct ifnet *ifp, struct mbuf *m);
144 static void oce_multiq_flush(struct ifnet *ifp);
146 /* Driver interrupt routines protypes */
147 static void oce_intr(void *arg, int pending);
148 static int oce_setup_intr(POCE_SOFTC sc);
149 static int oce_fast_isr(void *arg);
150 static int oce_alloc_intr(POCE_SOFTC sc, int vector,
151 void (*isr) (void *arg, int pending));
153 /* Media callbacks prototypes */
154 static void oce_media_status(struct ifnet *ifp, struct ifmediareq *req);
155 static int oce_media_change(struct ifnet *ifp);
157 /* Transmit routines prototypes */
158 static int oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index);
159 static void oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq);
160 static void oce_process_tx_completion(struct oce_wq *wq);
161 static int oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m,
164 /* Receive routines prototypes */
165 static int oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
166 static int oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
167 static void oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe);
168 static void oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq);
169 static uint16_t oce_rq_handler_lro(void *arg);
170 static void oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2);
171 static void oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2);
172 static void oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m);
174 /* Helper function prototypes in this file */
175 static int oce_attach_ifp(POCE_SOFTC sc);
176 static void oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
177 static void oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
178 static int oce_vid_config(POCE_SOFTC sc);
179 static void oce_mac_addr_set(POCE_SOFTC sc);
180 static int oce_handle_passthrough(struct ifnet *ifp, caddr_t data);
181 static void oce_local_timer(void *arg);
182 static void oce_if_deactivate(POCE_SOFTC sc);
183 static void oce_if_activate(POCE_SOFTC sc);
184 static void setup_max_queues_want(POCE_SOFTC sc);
185 static void update_queues_got(POCE_SOFTC sc);
186 static void process_link_state(POCE_SOFTC sc,
187 struct oce_async_cqe_link_state *acqe);
188 static int oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m);
189 static void oce_get_config(POCE_SOFTC sc);
190 static struct mbuf *oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete);
191 static void oce_read_env_variables(POCE_SOFTC sc);
195 #if defined(INET6) || defined(INET)
196 static int oce_init_lro(POCE_SOFTC sc);
197 static struct mbuf * oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp);
200 static device_method_t oce_dispatch[] = {
201 DEVMETHOD(device_probe, oce_probe),
202 DEVMETHOD(device_attach, oce_attach),
203 DEVMETHOD(device_detach, oce_detach),
204 DEVMETHOD(device_shutdown, oce_shutdown),
209 static driver_t oce_driver = {
214 static devclass_t oce_devclass;
218 const char component_revision[32] = {"///" COMPONENT_REVISION "///"};
220 /* Module capabilites and parameters */
221 uint32_t oce_max_rsp_handled = OCE_MAX_RSP_HANDLED;
222 uint32_t oce_enable_rss = OCE_MODCAP_RSS;
223 uint32_t oce_rq_buf_size = 2048;
225 TUNABLE_INT("hw.oce.max_rsp_handled", &oce_max_rsp_handled);
226 TUNABLE_INT("hw.oce.enable_rss", &oce_enable_rss);
229 /* Supported devices table */
230 static uint32_t supportedDevices[] = {
231 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE2,
232 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE3,
233 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_BE3,
234 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201,
235 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201_VF,
236 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_SH
240 DRIVER_MODULE(oce, pci, oce_driver, oce_devclass, 0, 0);
241 MODULE_PNP_INFO("W32:vendor/device", pci, oce, supportedDevices,
242 nitems(supportedDevices));
243 MODULE_DEPEND(oce, pci, 1, 1, 1);
244 MODULE_DEPEND(oce, ether, 1, 1, 1);
245 MODULE_VERSION(oce, 1);
248 POCE_SOFTC softc_head = NULL;
249 POCE_SOFTC softc_tail = NULL;
251 struct oce_rdma_if *oce_rdma_if = NULL;
253 /*****************************************************************************
254 * Driver entry points functions *
255 *****************************************************************************/
258 oce_probe(device_t dev)
266 sc = device_get_softc(dev);
267 bzero(sc, sizeof(OCE_SOFTC));
270 vendor = pci_get_vendor(dev);
271 device = pci_get_device(dev);
273 for (i = 0; i < (sizeof(supportedDevices) / sizeof(uint32_t)); i++) {
274 if (vendor == ((supportedDevices[i] >> 16) & 0xffff)) {
275 if (device == (supportedDevices[i] & 0xffff)) {
276 sprintf(str, "%s:%s", "Emulex CNA NIC function",
278 device_set_desc_copy(dev, str);
281 case PCI_PRODUCT_BE2:
282 sc->flags |= OCE_FLAGS_BE2;
284 case PCI_PRODUCT_BE3:
285 sc->flags |= OCE_FLAGS_BE3;
287 case PCI_PRODUCT_XE201:
288 case PCI_PRODUCT_XE201_VF:
289 sc->flags |= OCE_FLAGS_XE201;
292 sc->flags |= OCE_FLAGS_SH;
297 return BUS_PROBE_DEFAULT;
307 oce_attach(device_t dev)
312 sc = device_get_softc(dev);
314 rc = oce_hw_pci_alloc(sc);
318 sc->tx_ring_size = OCE_TX_RING_SIZE;
319 sc->rx_ring_size = OCE_RX_RING_SIZE;
320 /* receive fragment size should be multiple of 2K */
321 sc->rq_frag_size = ((oce_rq_buf_size / 2048) * 2048);
322 sc->flow_control = OCE_DEFAULT_FLOW_CONTROL;
323 sc->promisc = OCE_DEFAULT_PROMISCUOUS;
325 LOCK_CREATE(&sc->bmbx_lock, "Mailbox_lock");
326 LOCK_CREATE(&sc->dev_lock, "Device_lock");
328 /* initialise the hardware */
329 rc = oce_hw_init(sc);
333 oce_read_env_variables(sc);
337 setup_max_queues_want(sc);
339 rc = oce_setup_intr(sc);
343 rc = oce_queue_init_all(sc);
347 rc = oce_attach_ifp(sc);
351 #if defined(INET6) || defined(INET)
352 rc = oce_init_lro(sc);
357 rc = oce_hw_start(sc);
361 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
362 oce_add_vlan, sc, EVENTHANDLER_PRI_FIRST);
363 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
364 oce_del_vlan, sc, EVENTHANDLER_PRI_FIRST);
366 rc = oce_stats_init(sc);
372 callout_init(&sc->timer, CALLOUT_MPSAFE);
373 rc = callout_reset(&sc->timer, 2 * hz, oce_local_timer, sc);
378 if (softc_tail != NULL) {
379 softc_tail->next = sc;
388 callout_drain(&sc->timer);
392 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
394 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
395 oce_hw_intr_disable(sc);
397 #if defined(INET6) || defined(INET)
401 ether_ifdetach(sc->ifp);
404 oce_queue_release_all(sc);
408 oce_dma_free(sc, &sc->bsmbx);
411 LOCK_DESTROY(&sc->dev_lock);
412 LOCK_DESTROY(&sc->bmbx_lock);
419 oce_detach(device_t dev)
421 POCE_SOFTC sc = device_get_softc(dev);
422 POCE_SOFTC poce_sc_tmp, *ppoce_sc_tmp1, poce_sc_tmp2 = NULL;
424 poce_sc_tmp = softc_head;
425 ppoce_sc_tmp1 = &softc_head;
426 while (poce_sc_tmp != NULL) {
427 if (poce_sc_tmp == sc) {
428 *ppoce_sc_tmp1 = sc->next;
429 if (sc->next == NULL) {
430 softc_tail = poce_sc_tmp2;
434 poce_sc_tmp2 = poce_sc_tmp;
435 ppoce_sc_tmp1 = &poce_sc_tmp->next;
436 poce_sc_tmp = poce_sc_tmp->next;
440 oce_if_deactivate(sc);
441 UNLOCK(&sc->dev_lock);
443 callout_drain(&sc->timer);
445 if (sc->vlan_attach != NULL)
446 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
447 if (sc->vlan_detach != NULL)
448 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
450 ether_ifdetach(sc->ifp);
456 bus_generic_detach(dev);
463 oce_shutdown(device_t dev)
467 rc = oce_detach(dev);
474 oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
476 struct ifreq *ifr = (struct ifreq *)data;
477 POCE_SOFTC sc = ifp->if_softc;
484 rc = ifmedia_ioctl(ifp, ifr, &sc->media, command);
488 if (ifr->ifr_mtu > OCE_MAX_MTU)
491 ifp->if_mtu = ifr->ifr_mtu;
495 if (ifp->if_flags & IFF_UP) {
496 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
497 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
500 device_printf(sc->dev, "Interface Up\n");
504 sc->ifp->if_drv_flags &=
505 ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
506 oce_if_deactivate(sc);
508 UNLOCK(&sc->dev_lock);
510 device_printf(sc->dev, "Interface Down\n");
513 if ((ifp->if_flags & IFF_PROMISC) && !sc->promisc) {
514 if (!oce_rxf_set_promiscuous(sc, (1 | (1 << 1))))
516 } else if (!(ifp->if_flags & IFF_PROMISC) && sc->promisc) {
517 if (!oce_rxf_set_promiscuous(sc, 0))
525 rc = oce_hw_update_multicast(sc);
527 device_printf(sc->dev,
528 "Update multicast address failed\n");
532 u = ifr->ifr_reqcap ^ ifp->if_capenable;
534 if (u & IFCAP_TXCSUM) {
535 ifp->if_capenable ^= IFCAP_TXCSUM;
536 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
538 if (IFCAP_TSO & ifp->if_capenable &&
539 !(IFCAP_TXCSUM & ifp->if_capenable)) {
540 ifp->if_capenable &= ~IFCAP_TSO;
541 ifp->if_hwassist &= ~CSUM_TSO;
543 "TSO disabled due to -txcsum.\n");
547 if (u & IFCAP_RXCSUM)
548 ifp->if_capenable ^= IFCAP_RXCSUM;
550 if (u & IFCAP_TSO4) {
551 ifp->if_capenable ^= IFCAP_TSO4;
553 if (IFCAP_TSO & ifp->if_capenable) {
554 if (IFCAP_TXCSUM & ifp->if_capenable)
555 ifp->if_hwassist |= CSUM_TSO;
557 ifp->if_capenable &= ~IFCAP_TSO;
558 ifp->if_hwassist &= ~CSUM_TSO;
560 "Enable txcsum first.\n");
564 ifp->if_hwassist &= ~CSUM_TSO;
567 if (u & IFCAP_VLAN_HWTAGGING)
568 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
570 if (u & IFCAP_VLAN_HWFILTER) {
571 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
574 #if defined(INET6) || defined(INET)
576 ifp->if_capenable ^= IFCAP_LRO;
577 if(sc->enable_hwlro) {
578 if(ifp->if_capenable & IFCAP_LRO) {
579 rc = oce_mbox_nic_set_iface_lro_config(sc, 1);
581 rc = oce_mbox_nic_set_iface_lro_config(sc, 0);
590 rc = oce_handle_passthrough(ifp, data);
593 rc = ether_ioctl(ifp, command, data);
608 if (sc->ifp->if_flags & IFF_UP) {
609 oce_if_deactivate(sc);
613 UNLOCK(&sc->dev_lock);
619 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
621 POCE_SOFTC sc = ifp->if_softc;
622 struct oce_wq *wq = NULL;
626 if (!sc->link_status)
629 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
630 queue_index = m->m_pkthdr.flowid % sc->nwqs;
632 wq = sc->wq[queue_index];
635 status = oce_multiq_transmit(ifp, m, wq);
636 UNLOCK(&wq->tx_lock);
644 oce_multiq_flush(struct ifnet *ifp)
646 POCE_SOFTC sc = ifp->if_softc;
650 for (i = 0; i < sc->nwqs; i++) {
651 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
659 /*****************************************************************************
660 * Driver interrupt routines functions *
661 *****************************************************************************/
664 oce_intr(void *arg, int pending)
667 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
668 POCE_SOFTC sc = ii->sc;
669 struct oce_eq *eq = ii->eq;
671 struct oce_cq *cq = NULL;
675 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
676 BUS_DMASYNC_POSTWRITE);
678 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
682 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
683 BUS_DMASYNC_POSTWRITE);
684 RING_GET(eq->ring, 1);
690 goto eq_arm; /* Spurious */
692 /* Clear EQ entries, but dont arm */
693 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
695 /* Process TX, RX and MCC. But dont arm CQ*/
696 for (i = 0; i < eq->cq_valid; i++) {
698 (*cq->cq_handler)(cq->cb_arg);
701 /* Arm all cqs connected to this EQ */
702 for (i = 0; i < eq->cq_valid; i++) {
704 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
708 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
715 oce_setup_intr(POCE_SOFTC sc)
717 int rc = 0, use_intx = 0;
718 int vector = 0, req_vectors = 0;
719 int tot_req_vectors, tot_vectors;
721 if (is_rss_enabled(sc))
722 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
726 tot_req_vectors = req_vectors;
727 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
728 if (req_vectors > 1) {
729 tot_req_vectors += OCE_RDMA_VECTORS;
730 sc->roce_intr_count = OCE_RDMA_VECTORS;
734 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
735 sc->intr_count = req_vectors;
736 tot_vectors = tot_req_vectors;
737 rc = pci_alloc_msix(sc->dev, &tot_vectors);
740 pci_release_msi(sc->dev);
742 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
743 if (tot_vectors < tot_req_vectors) {
744 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
745 sc->roce_intr_count = (tot_vectors / 2);
747 sc->intr_count = tot_vectors - sc->roce_intr_count;
750 sc->intr_count = tot_vectors;
752 sc->flags |= OCE_FLAGS_USING_MSIX;
760 /* Scale number of queues based on intr we got */
761 update_queues_got(sc);
764 device_printf(sc->dev, "Using legacy interrupt\n");
765 rc = oce_alloc_intr(sc, vector, oce_intr);
769 for (; vector < sc->intr_count; vector++) {
770 rc = oce_alloc_intr(sc, vector, oce_intr);
784 oce_fast_isr(void *arg)
786 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
787 POCE_SOFTC sc = ii->sc;
792 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
794 taskqueue_enqueue(ii->tq, &ii->task);
798 return FILTER_HANDLED;
803 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
805 POCE_INTR_INFO ii = &sc->intrs[vector];
808 if (vector >= OCE_MAX_EQ)
811 /* Set the resource id for the interrupt.
812 * MSIx is vector + 1 for the resource id,
813 * INTx is 0 for the resource id.
815 if (sc->flags & OCE_FLAGS_USING_MSIX)
819 ii->intr_res = bus_alloc_resource_any(sc->dev,
821 &rr, RF_ACTIVE|RF_SHAREABLE);
823 if (ii->intr_res == NULL) {
824 device_printf(sc->dev,
825 "Could not allocate interrupt\n");
830 TASK_INIT(&ii->task, 0, isr, ii);
832 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
833 ii->tq = taskqueue_create_fast(ii->task_name,
835 taskqueue_thread_enqueue,
837 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
838 device_get_nameunit(sc->dev));
841 rc = bus_setup_intr(sc->dev,
844 oce_fast_isr, NULL, ii, &ii->tag);
851 oce_intr_free(POCE_SOFTC sc)
855 for (i = 0; i < sc->intr_count; i++) {
857 if (sc->intrs[i].tag != NULL)
858 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
860 if (sc->intrs[i].tq != NULL)
861 taskqueue_free(sc->intrs[i].tq);
863 if (sc->intrs[i].intr_res != NULL)
864 bus_release_resource(sc->dev, SYS_RES_IRQ,
866 sc->intrs[i].intr_res);
867 sc->intrs[i].tag = NULL;
868 sc->intrs[i].intr_res = NULL;
871 if (sc->flags & OCE_FLAGS_USING_MSIX)
872 pci_release_msi(sc->dev);
878 /******************************************************************************
879 * Media callbacks functions *
880 ******************************************************************************/
883 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
885 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
888 req->ifm_status = IFM_AVALID;
889 req->ifm_active = IFM_ETHER;
891 if (sc->link_status == 1)
892 req->ifm_status |= IFM_ACTIVE;
896 switch (sc->link_speed) {
897 case 1: /* 10 Mbps */
898 req->ifm_active |= IFM_10_T | IFM_FDX;
901 case 2: /* 100 Mbps */
902 req->ifm_active |= IFM_100_TX | IFM_FDX;
906 req->ifm_active |= IFM_1000_T | IFM_FDX;
909 case 4: /* 10 Gbps */
910 req->ifm_active |= IFM_10G_SR | IFM_FDX;
913 case 5: /* 20 Gbps */
914 req->ifm_active |= IFM_10G_SR | IFM_FDX;
917 case 6: /* 25 Gbps */
918 req->ifm_active |= IFM_10G_SR | IFM_FDX;
921 case 7: /* 40 Gbps */
922 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
935 oce_media_change(struct ifnet *ifp)
941 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
942 struct mbuf *m, boolean_t *os2bmc,
945 struct ether_header *eh = NULL;
947 eh = mtod(m, struct ether_header *);
949 if (!is_os2bmc_enabled(sc) || *os2bmc) {
953 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
956 if (is_mc_allowed_on_bmc(sc, eh) ||
957 is_bc_allowed_on_bmc(sc, eh) ||
958 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
963 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
964 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
965 uint8_t nexthdr = ip6->ip6_nxt;
966 if (nexthdr == IPPROTO_ICMPV6) {
967 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
968 switch (icmp6->icmp6_type) {
969 case ND_ROUTER_ADVERT:
970 *os2bmc = is_ipv6_ra_filt_enabled(sc);
972 case ND_NEIGHBOR_ADVERT:
973 *os2bmc = is_ipv6_na_filt_enabled(sc);
981 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
982 struct ip *ip = mtod(m, struct ip *);
983 int iphlen = ip->ip_hl << 2;
984 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
985 switch (uh->uh_dport) {
986 case DHCP_CLIENT_PORT:
987 *os2bmc = is_dhcp_client_filt_enabled(sc);
989 case DHCP_SERVER_PORT:
990 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
994 *os2bmc = is_nbios_filt_enabled(sc);
996 case DHCPV6_RAS_PORT:
997 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1005 *m_new = m_dup(m, M_NOWAIT);
1010 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1016 /*****************************************************************************
1017 * Transmit routines functions *
1018 *****************************************************************************/
1021 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1023 int rc = 0, i, retry_cnt = 0;
1024 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1025 struct mbuf *m, *m_temp, *m_new = NULL;
1026 struct oce_wq *wq = sc->wq[wq_index];
1027 struct oce_packet_desc *pd;
1028 struct oce_nic_hdr_wqe *nichdr;
1029 struct oce_nic_frag_wqe *nicfrag;
1030 struct ether_header *eh = NULL;
1033 boolean_t complete = TRUE;
1034 boolean_t os2bmc = FALSE;
1040 if (!(m->m_flags & M_PKTHDR)) {
1045 /* Don't allow non-TSO packets longer than MTU */
1046 if (!is_tso_pkt(m)) {
1047 eh = mtod(m, struct ether_header *);
1048 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1052 if(oce_tx_asic_stall_verify(sc, m)) {
1053 m = oce_insert_vlan_tag(sc, m, &complete);
1055 device_printf(sc->dev, "Insertion unsuccessful\n");
1061 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1062 * may cause a transmit stall on that port. So the work-around is to
1063 * pad short packets (<= 32 bytes) to a 36-byte length.
1065 if(IS_SH(sc) || IS_XE201(sc) ) {
1066 if(m->m_pkthdr.len <= 32) {
1068 bzero((void *)buf, 36);
1069 m_append(m, (36 - m->m_pkthdr.len), buf);
1074 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1075 /* consolidate packet buffers for TSO/LSO segment offload */
1076 #if defined(INET6) || defined(INET)
1077 m = oce_tso_setup(sc, mpp);
1088 pd = &wq->pckts[wq->pkt_desc_head];
1091 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1093 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1095 num_wqes = pd->nsegs + 1;
1096 if (IS_BE(sc) || IS_SH(sc)) {
1097 /*Dummy required only for BE3.*/
1101 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1102 bus_dmamap_unload(wq->tag, pd->map);
1105 atomic_store_rel_int(&wq->pkt_desc_head,
1106 (wq->pkt_desc_head + 1) % \
1107 OCE_WQ_PACKET_ARRAY_SIZE);
1108 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1112 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1113 nichdr->u0.dw[0] = 0;
1114 nichdr->u0.dw[1] = 0;
1115 nichdr->u0.dw[2] = 0;
1116 nichdr->u0.dw[3] = 0;
1118 nichdr->u0.s.complete = complete;
1119 nichdr->u0.s.mgmt = os2bmc;
1120 nichdr->u0.s.event = 1;
1121 nichdr->u0.s.crc = 1;
1122 nichdr->u0.s.forward = 0;
1123 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1124 nichdr->u0.s.udpcs =
1125 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1126 nichdr->u0.s.tcpcs =
1127 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1128 nichdr->u0.s.num_wqe = num_wqes;
1129 nichdr->u0.s.total_length = m->m_pkthdr.len;
1131 if (m->m_flags & M_VLANTAG) {
1132 nichdr->u0.s.vlan = 1; /*Vlan present*/
1133 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1136 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1137 if (m->m_pkthdr.tso_segsz) {
1138 nichdr->u0.s.lso = 1;
1139 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1141 if (!IS_BE(sc) || !IS_SH(sc))
1142 nichdr->u0.s.ipcs = 1;
1145 RING_PUT(wq->ring, 1);
1146 atomic_add_int(&wq->ring->num_used, 1);
1148 for (i = 0; i < pd->nsegs; i++) {
1150 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1151 struct oce_nic_frag_wqe);
1152 nicfrag->u0.s.rsvd0 = 0;
1153 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1154 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1155 nicfrag->u0.s.frag_len = segs[i].ds_len;
1156 pd->wqe_idx = wq->ring->pidx;
1157 RING_PUT(wq->ring, 1);
1158 atomic_add_int(&wq->ring->num_used, 1);
1160 if (num_wqes > (pd->nsegs + 1)) {
1162 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1163 struct oce_nic_frag_wqe);
1164 nicfrag->u0.dw[0] = 0;
1165 nicfrag->u0.dw[1] = 0;
1166 nicfrag->u0.dw[2] = 0;
1167 nicfrag->u0.dw[3] = 0;
1168 pd->wqe_idx = wq->ring->pidx;
1169 RING_PUT(wq->ring, 1);
1170 atomic_add_int(&wq->ring->num_used, 1);
1174 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1175 wq->tx_stats.tx_reqs++;
1176 wq->tx_stats.tx_wrbs += num_wqes;
1177 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1178 wq->tx_stats.tx_pkts++;
1180 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1181 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1182 reg_value = (num_wqes << 16) | wq->wq_id;
1184 /* if os2bmc is not enabled or if the pkt is already tagged as
1187 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1189 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1191 } else if (rc == EFBIG) {
1192 if (retry_cnt == 0) {
1193 m_temp = m_defrag(m, M_NOWAIT);
1198 retry_cnt = retry_cnt + 1;
1202 } else if (rc == ENOMEM)
1222 oce_process_tx_completion(struct oce_wq *wq)
1224 struct oce_packet_desc *pd;
1225 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1228 pd = &wq->pckts[wq->pkt_desc_tail];
1229 atomic_store_rel_int(&wq->pkt_desc_tail,
1230 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1231 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1232 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1233 bus_dmamap_unload(wq->tag, pd->map);
1240 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1241 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1242 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1243 oce_tx_restart(sc, wq);
1250 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1253 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1256 #if __FreeBSD_version >= 800000
1257 if (!drbr_empty(sc->ifp, wq->br))
1259 if (!IFQ_DRV_IS_EMPTY(&sc->ifp->if_snd))
1261 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1266 #if defined(INET6) || defined(INET)
1267 static struct mbuf *
1268 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1275 struct ip6_hdr *ip6;
1277 struct ether_vlan_header *eh;
1280 int total_len = 0, ehdrlen = 0;
1284 if (M_WRITABLE(m) == 0) {
1285 m = m_dup(*mpp, M_NOWAIT);
1292 eh = mtod(m, struct ether_vlan_header *);
1293 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1294 etype = ntohs(eh->evl_proto);
1295 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1297 etype = ntohs(eh->evl_encap_proto);
1298 ehdrlen = ETHER_HDR_LEN;
1304 ip = (struct ip *)(m->m_data + ehdrlen);
1305 if (ip->ip_p != IPPROTO_TCP)
1307 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1309 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1313 case ETHERTYPE_IPV6:
1314 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1315 if (ip6->ip6_nxt != IPPROTO_TCP)
1317 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1319 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1326 m = m_pullup(m, total_len);
1333 #endif /* INET6 || INET */
1336 oce_tx_task(void *arg, int npending)
1338 struct oce_wq *wq = arg;
1339 POCE_SOFTC sc = wq->parent;
1340 struct ifnet *ifp = sc->ifp;
1343 #if __FreeBSD_version >= 800000
1345 rc = oce_multiq_transmit(ifp, NULL, wq);
1347 device_printf(sc->dev,
1348 "TX[%d] restart failed\n", wq->queue_index);
1350 UNLOCK(&wq->tx_lock);
1359 oce_start(struct ifnet *ifp)
1361 POCE_SOFTC sc = ifp->if_softc;
1364 int def_q = 0; /* Defualt tx queue is 0*/
1366 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1370 if (!sc->link_status)
1374 IF_DEQUEUE(&sc->ifp->if_snd, m);
1378 LOCK(&sc->wq[def_q]->tx_lock);
1379 rc = oce_tx(sc, &m, def_q);
1380 UNLOCK(&sc->wq[def_q]->tx_lock);
1383 sc->wq[def_q]->tx_stats.tx_stops ++;
1384 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1385 IFQ_DRV_PREPEND(&ifp->if_snd, m);
1391 ETHER_BPF_MTAP(ifp, m);
1399 /* Handle the Completion Queue for transmit */
1401 oce_wq_handler(void *arg)
1403 struct oce_wq *wq = (struct oce_wq *)arg;
1404 POCE_SOFTC sc = wq->parent;
1405 struct oce_cq *cq = wq->cq;
1406 struct oce_nic_tx_cqe *cqe;
1409 LOCK(&wq->tx_compl_lock);
1410 bus_dmamap_sync(cq->ring->dma.tag,
1411 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1412 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1413 while (cqe->u0.dw[3]) {
1414 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe));
1416 wq->ring->cidx = cqe->u0.s.wqe_index + 1;
1417 if (wq->ring->cidx >= wq->ring->num_items)
1418 wq->ring->cidx -= wq->ring->num_items;
1420 oce_process_tx_completion(wq);
1421 wq->tx_stats.tx_compl++;
1423 RING_GET(cq->ring, 1);
1424 bus_dmamap_sync(cq->ring->dma.tag,
1425 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1427 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1432 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
1434 UNLOCK(&wq->tx_compl_lock);
1440 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq)
1442 POCE_SOFTC sc = ifp->if_softc;
1443 int status = 0, queue_index = 0;
1444 struct mbuf *next = NULL;
1445 struct buf_ring *br = NULL;
1448 queue_index = wq->queue_index;
1450 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1453 status = drbr_enqueue(ifp, br, m);
1458 if ((status = drbr_enqueue(ifp, br, m)) != 0)
1461 while ((next = drbr_peek(ifp, br)) != NULL) {
1462 if (oce_tx(sc, &next, queue_index)) {
1464 drbr_advance(ifp, br);
1466 drbr_putback(ifp, br, next);
1467 wq->tx_stats.tx_stops ++;
1468 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1472 drbr_advance(ifp, br);
1473 if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
1474 if (next->m_flags & M_MCAST)
1475 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1476 ETHER_BPF_MTAP(ifp, next);
1485 /*****************************************************************************
1486 * Receive routines functions *
1487 *****************************************************************************/
1490 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2)
1493 struct ether_header *eh = NULL;
1494 struct tcphdr *tcp_hdr = NULL;
1495 struct ip *ip4_hdr = NULL;
1496 struct ip6_hdr *ip6 = NULL;
1497 uint32_t payload_len = 0;
1499 eh = mtod(m, struct ether_header *);
1500 /* correct IP header */
1501 if(!cqe2->ipv6_frame) {
1502 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header));
1503 ip4_hdr->ip_ttl = cqe2->frame_lifespan;
1504 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header));
1505 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip));
1507 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header));
1508 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan;
1509 payload_len = cqe2->coalesced_size - sizeof(struct ether_header)
1510 - sizeof(struct ip6_hdr);
1511 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len);
1512 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr));
1515 /* correct tcp header */
1516 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num);
1518 tcp_hdr->th_flags |= TH_PUSH;
1520 tcp_hdr->th_win = htons(cqe2->tcp_window);
1521 tcp_hdr->th_sum = 0xffff;
1523 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2);
1524 *p = cqe1->tcp_timestamp_val;
1525 *(p+1) = cqe1->tcp_timestamp_ecr;
1532 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m)
1534 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1535 uint32_t i = 0, frag_len = 0;
1536 uint32_t len = cqe_info->pkt_size;
1537 struct oce_packet_desc *pd;
1538 struct mbuf *tail = NULL;
1540 for (i = 0; i < cqe_info->num_frags; i++) {
1541 if (rq->ring->cidx == rq->ring->pidx) {
1542 device_printf(sc->dev,
1543 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n");
1546 pd = &rq->pckts[rq->ring->cidx];
1548 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1549 bus_dmamap_unload(rq->tag, pd->map);
1550 RING_GET(rq->ring, 1);
1553 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len;
1554 pd->mbuf->m_len = frag_len;
1557 /* additional fragments */
1558 pd->mbuf->m_flags &= ~M_PKTHDR;
1559 tail->m_next = pd->mbuf;
1561 tail->m_nextpkt = NULL;
1564 /* first fragment, fill out much of the packet header */
1565 pd->mbuf->m_pkthdr.len = len;
1567 pd->mbuf->m_nextpkt = NULL;
1568 pd->mbuf->m_pkthdr.csum_flags = 0;
1569 if (IF_CSUM_ENABLED(sc)) {
1570 if (cqe_info->l4_cksum_pass) {
1571 if(!cqe_info->ipv6_frame) { /* IPV4 */
1572 pd->mbuf->m_pkthdr.csum_flags |=
1573 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1574 }else { /* IPV6 frame */
1576 pd->mbuf->m_pkthdr.csum_flags |=
1577 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1580 pd->mbuf->m_pkthdr.csum_data = 0xffff;
1582 if (cqe_info->ip_cksum_pass) {
1583 pd->mbuf->m_pkthdr.csum_flags |=
1584 (CSUM_IP_CHECKED|CSUM_IP_VALID);
1587 *m = tail = pd->mbuf;
1597 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2)
1599 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1600 struct nic_hwlro_cqe_part1 *cqe1 = NULL;
1601 struct mbuf *m = NULL;
1602 struct oce_common_cqe_info cq_info;
1606 cq_info.pkt_size = cqe->pkt_size;
1607 cq_info.vtag = cqe->vlan_tag;
1608 cq_info.l4_cksum_pass = cqe->l4_cksum_pass;
1609 cq_info.ip_cksum_pass = cqe->ip_cksum_pass;
1610 cq_info.ipv6_frame = cqe->ipv6_frame;
1611 cq_info.vtp = cqe->vtp;
1612 cq_info.qnq = cqe->qnq;
1614 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe;
1615 cq_info.pkt_size = cqe2->coalesced_size;
1616 cq_info.vtag = cqe2->vlan_tag;
1617 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass;
1618 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass;
1619 cq_info.ipv6_frame = cqe2->ipv6_frame;
1620 cq_info.vtp = cqe2->vtp;
1621 cq_info.qnq = cqe1->qnq;
1624 cq_info.vtag = BSWAP_16(cq_info.vtag);
1626 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size;
1627 if(cq_info.pkt_size % rq->cfg.frag_size)
1628 cq_info.num_frags++;
1630 oce_rx_mbuf_chain(rq, &cq_info, &m);
1634 //assert(cqe2->valid != 0);
1636 //assert(cqe2->cqe_type != 2);
1637 oce_correct_header(m, cqe1, cqe2);
1640 m->m_pkthdr.rcvif = sc->ifp;
1641 #if __FreeBSD_version >= 800000
1642 if (rq->queue_index)
1643 m->m_pkthdr.flowid = (rq->queue_index - 1);
1645 m->m_pkthdr.flowid = rq->queue_index;
1646 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1648 /* This deternies if vlan tag is Valid */
1650 if (sc->function_mode & FNM_FLEX10_MODE) {
1651 /* FLEX10. If QnQ is not set, neglect VLAN */
1653 m->m_pkthdr.ether_vtag = cq_info.vtag;
1654 m->m_flags |= M_VLANTAG;
1656 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1657 /* In UMC mode generally pvid will be striped by
1658 hw. But in some cases we have seen it comes
1659 with pvid. So if pvid == vlan, neglect vlan.
1661 m->m_pkthdr.ether_vtag = cq_info.vtag;
1662 m->m_flags |= M_VLANTAG;
1665 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1667 (*sc->ifp->if_input) (sc->ifp, m);
1669 /* Update rx stats per queue */
1670 rq->rx_stats.rx_pkts++;
1671 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1672 rq->rx_stats.rx_frags += cq_info.num_frags;
1673 rq->rx_stats.rx_ucast_pkts++;
1679 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1681 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1683 struct mbuf *m = NULL;
1684 struct oce_common_cqe_info cq_info;
1687 /* Is it a flush compl that has no data */
1688 if(!cqe->u0.s.num_fragments)
1691 len = cqe->u0.s.pkt_size;
1693 /*partial DMA workaround for Lancer*/
1694 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1698 if (!oce_cqe_portid_valid(sc, cqe)) {
1699 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1703 /* Get vlan_tag value */
1704 if(IS_BE(sc) || IS_SH(sc))
1705 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1707 vtag = cqe->u0.s.vlan_tag;
1709 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1710 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1711 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1712 cq_info.num_frags = cqe->u0.s.num_fragments;
1713 cq_info.pkt_size = cqe->u0.s.pkt_size;
1715 oce_rx_mbuf_chain(rq, &cq_info, &m);
1718 m->m_pkthdr.rcvif = sc->ifp;
1719 #if __FreeBSD_version >= 800000
1720 if (rq->queue_index)
1721 m->m_pkthdr.flowid = (rq->queue_index - 1);
1723 m->m_pkthdr.flowid = rq->queue_index;
1724 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1726 /* This deternies if vlan tag is Valid */
1727 if (oce_cqe_vtp_valid(sc, cqe)) {
1728 if (sc->function_mode & FNM_FLEX10_MODE) {
1729 /* FLEX10. If QnQ is not set, neglect VLAN */
1730 if (cqe->u0.s.qnq) {
1731 m->m_pkthdr.ether_vtag = vtag;
1732 m->m_flags |= M_VLANTAG;
1734 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1735 /* In UMC mode generally pvid will be striped by
1736 hw. But in some cases we have seen it comes
1737 with pvid. So if pvid == vlan, neglect vlan.
1739 m->m_pkthdr.ether_vtag = vtag;
1740 m->m_flags |= M_VLANTAG;
1744 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1745 #if defined(INET6) || defined(INET)
1746 /* Try to queue to LRO */
1747 if (IF_LRO_ENABLED(sc) &&
1748 (cqe->u0.s.ip_cksum_pass) &&
1749 (cqe->u0.s.l4_cksum_pass) &&
1750 (!cqe->u0.s.ip_ver) &&
1751 (rq->lro.lro_cnt != 0)) {
1753 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1754 rq->lro_pkts_queued ++;
1757 /* If LRO posting fails then try to post to STACK */
1761 (*sc->ifp->if_input) (sc->ifp, m);
1762 #if defined(INET6) || defined(INET)
1765 /* Update rx stats per queue */
1766 rq->rx_stats.rx_pkts++;
1767 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1768 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1769 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1770 rq->rx_stats.rx_mcast_pkts++;
1771 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1772 rq->rx_stats.rx_ucast_pkts++;
1780 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1783 struct oce_packet_desc *pd;
1784 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1786 for (i = 0; i < num_frags; i++) {
1787 if (rq->ring->cidx == rq->ring->pidx) {
1788 device_printf(sc->dev,
1789 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1792 pd = &rq->pckts[rq->ring->cidx];
1793 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1794 bus_dmamap_unload(rq->tag, pd->map);
1795 if (pd->mbuf != NULL) {
1800 RING_GET(rq->ring, 1);
1807 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1809 struct oce_nic_rx_cqe_v1 *cqe_v1;
1812 if (sc->be3_native) {
1813 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1814 vtp = cqe_v1->u0.s.vlan_tag_present;
1816 vtp = cqe->u0.s.vlan_tag_present;
1824 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1826 struct oce_nic_rx_cqe_v1 *cqe_v1;
1829 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1830 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1831 port_id = cqe_v1->u0.s.port;
1832 if (sc->port_id != port_id)
1835 ;/* For BE3 legacy and Lancer this is dummy */
1841 #if defined(INET6) || defined(INET)
1843 oce_rx_flush_lro(struct oce_rq *rq)
1845 struct lro_ctrl *lro = &rq->lro;
1846 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1848 if (!IF_LRO_ENABLED(sc))
1851 tcp_lro_flush_all(lro);
1852 rq->lro_pkts_queued = 0;
1859 oce_init_lro(POCE_SOFTC sc)
1861 struct lro_ctrl *lro = NULL;
1864 for (i = 0; i < sc->nrqs; i++) {
1865 lro = &sc->rq[i]->lro;
1866 rc = tcp_lro_init(lro);
1868 device_printf(sc->dev, "LRO init failed\n");
1879 oce_free_lro(POCE_SOFTC sc)
1881 struct lro_ctrl *lro = NULL;
1884 for (i = 0; i < sc->nrqs; i++) {
1885 lro = &sc->rq[i]->lro;
1893 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1895 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1897 struct oce_packet_desc *pd;
1898 bus_dma_segment_t segs[6];
1899 int nsegs, added = 0;
1900 struct oce_nic_rqe *rqe;
1901 pd_rxulp_db_t rxdb_reg;
1903 uint32_t oce_max_rq_posts = 64;
1905 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1906 for (i = 0; i < count; i++) {
1907 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1909 pd = &rq->pckts[rq->ring->pidx];
1910 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1911 if (pd->mbuf == NULL) {
1912 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1915 pd->mbuf->m_nextpkt = NULL;
1917 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1919 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1922 segs, &nsegs, BUS_DMA_NOWAIT);
1925 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1934 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1936 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1937 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1938 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1939 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1940 RING_PUT(rq->ring, 1);
1944 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1946 for (i = added / oce_max_rq_posts; i > 0; i--) {
1947 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1948 rxdb_reg.bits.qid = rq->rq_id;
1950 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1951 val |= oce_max_rq_posts << 16;
1952 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1954 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1956 added -= oce_max_rq_posts;
1959 rxdb_reg.bits.qid = rq->rq_id;
1960 rxdb_reg.bits.num_posted = added;
1962 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1964 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1966 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1975 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
1978 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
1979 if(!sc->enable_hwlro) {
1980 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
1981 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
1983 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
1984 oce_alloc_rx_bufs(rq, 64);
1992 oce_rq_handler_lro(void *arg)
1994 struct oce_rq *rq = (struct oce_rq *)arg;
1995 struct oce_cq *cq = rq->cq;
1996 POCE_SOFTC sc = rq->parent;
1997 struct nic_hwlro_singleton_cqe *cqe;
1998 struct nic_hwlro_cqe_part2 *cqe2;
2002 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2003 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2004 while (cqe->valid) {
2005 if(cqe->cqe_type == 0) { /* singleton cqe */
2006 /* we should not get singleton cqe after cqe1 on same rq */
2007 if(rq->cqe_firstpart != NULL) {
2008 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2009 goto exit_rq_handler_lro;
2011 if(cqe->error != 0) {
2012 rq->rx_stats.rxcp_err++;
2013 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2015 oce_rx_lro(rq, cqe, NULL);
2016 rq->rx_stats.rx_compl++;
2018 RING_GET(cq->ring, 1);
2020 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2022 }else if(cqe->cqe_type == 0x1) { /* first part */
2023 /* we should not get cqe1 after cqe1 on same rq */
2024 if(rq->cqe_firstpart != NULL) {
2025 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2026 goto exit_rq_handler_lro;
2028 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2029 RING_GET(cq->ring, 1);
2030 }else if(cqe->cqe_type == 0x2) { /* second part */
2031 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2032 if(cqe2->error != 0) {
2033 rq->rx_stats.rxcp_err++;
2034 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2036 /* We should not get cqe2 without cqe1 */
2037 if(rq->cqe_firstpart == NULL) {
2038 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2039 goto exit_rq_handler_lro;
2041 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2043 rq->rx_stats.rx_compl++;
2044 rq->cqe_firstpart->valid = 0;
2046 rq->cqe_firstpart = NULL;
2048 RING_GET(cq->ring, 1);
2050 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2054 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2055 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2057 oce_check_rx_bufs(sc, num_cqes, rq);
2058 exit_rq_handler_lro:
2059 UNLOCK(&rq->rx_lock);
2063 /* Handle the Completion Queue for receive */
2065 oce_rq_handler(void *arg)
2067 struct oce_rq *rq = (struct oce_rq *)arg;
2068 struct oce_cq *cq = rq->cq;
2069 POCE_SOFTC sc = rq->parent;
2070 struct oce_nic_rx_cqe *cqe;
2074 oce_rq_handler_lro(arg);
2078 bus_dmamap_sync(cq->ring->dma.tag,
2079 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2080 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2081 while (cqe->u0.dw[2]) {
2082 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2084 if (cqe->u0.s.error == 0) {
2087 rq->rx_stats.rxcp_err++;
2088 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2089 /* Post L3/L4 errors to stack.*/
2092 rq->rx_stats.rx_compl++;
2095 #if defined(INET6) || defined(INET)
2096 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2097 oce_rx_flush_lro(rq);
2101 RING_GET(cq->ring, 1);
2102 bus_dmamap_sync(cq->ring->dma.tag,
2103 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2105 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2107 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2111 #if defined(INET6) || defined(INET)
2112 if (IF_LRO_ENABLED(sc))
2113 oce_rx_flush_lro(rq);
2116 oce_check_rx_bufs(sc, num_cqes, rq);
2117 UNLOCK(&rq->rx_lock);
2125 /*****************************************************************************
2126 * Helper function prototypes in this file *
2127 *****************************************************************************/
2130 oce_attach_ifp(POCE_SOFTC sc)
2133 sc->ifp = if_alloc(IFT_ETHER);
2137 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2138 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2139 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2141 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2142 sc->ifp->if_ioctl = oce_ioctl;
2143 sc->ifp->if_start = oce_start;
2144 sc->ifp->if_init = oce_init;
2145 sc->ifp->if_mtu = ETHERMTU;
2146 sc->ifp->if_softc = sc;
2147 #if __FreeBSD_version >= 800000
2148 sc->ifp->if_transmit = oce_multiq_start;
2149 sc->ifp->if_qflush = oce_multiq_flush;
2152 if_initname(sc->ifp,
2153 device_get_name(sc->dev), device_get_unit(sc->dev));
2155 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2156 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2157 IFQ_SET_READY(&sc->ifp->if_snd);
2159 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2160 sc->ifp->if_hwassist |= CSUM_TSO;
2161 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2163 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2164 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2165 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2167 #if defined(INET6) || defined(INET)
2168 sc->ifp->if_capabilities |= IFCAP_TSO;
2169 sc->ifp->if_capabilities |= IFCAP_LRO;
2170 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2173 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2174 sc->ifp->if_baudrate = IF_Gbps(10);
2176 #if __FreeBSD_version >= 1000000
2177 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2178 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2179 sc->ifp->if_hw_tsomaxsegsize = 4096;
2182 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2189 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2191 POCE_SOFTC sc = ifp->if_softc;
2193 if (ifp->if_softc != arg)
2195 if ((vtag == 0) || (vtag > 4095))
2198 sc->vlan_tag[vtag] = 1;
2200 if (sc->vlans_added <= (sc->max_vlans + 1))
2206 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2208 POCE_SOFTC sc = ifp->if_softc;
2210 if (ifp->if_softc != arg)
2212 if ((vtag == 0) || (vtag > 4095))
2215 sc->vlan_tag[vtag] = 0;
2222 * A max of 64 vlans can be configured in BE. If the user configures
2223 * more, place the card in vlan promiscuous mode.
2226 oce_vid_config(POCE_SOFTC sc)
2228 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2229 uint16_t ntags = 0, i;
2232 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2233 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2234 for (i = 0; i < MAX_VLANS; i++) {
2235 if (sc->vlan_tag[i]) {
2236 vtags[ntags].vtag = i;
2241 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2242 vtags, ntags, 1, 0);
2244 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2251 oce_mac_addr_set(POCE_SOFTC sc)
2253 uint32_t old_pmac_id = sc->pmac_id;
2257 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2258 sc->macaddr.size_of_struct);
2262 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2263 sc->if_id, &sc->pmac_id);
2265 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2266 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2267 sc->macaddr.size_of_struct);
2270 device_printf(sc->dev, "Failed update macaddress\n");
2276 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2278 POCE_SOFTC sc = ifp->if_softc;
2279 struct ifreq *ifr = (struct ifreq *)data;
2281 char cookie[32] = {0};
2282 void *priv_data = ifr_data_get_ptr(ifr);
2286 OCE_DMA_MEM dma_mem;
2287 struct mbx_common_get_cntl_attr *fw_cmd;
2289 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2292 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2295 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2296 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2299 req_size = le32toh(req.u0.req.request_length);
2300 if (req_size > 65536)
2303 req_size += sizeof(struct mbx_hdr);
2304 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2308 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2313 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2319 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2323 firmware is filling all the attributes for this ioctl except
2324 the driver version..so fill it
2326 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2327 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2328 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2329 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2333 oce_dma_free(sc, &dma_mem);
2339 oce_eqd_set_periodic(POCE_SOFTC sc)
2341 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2342 struct oce_aic_obj *aic;
2344 uint64_t now = 0, delta;
2345 int eqd, i, num = 0;
2346 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2350 #define ticks_to_msecs(t) (1000 * (t) / hz)
2352 for (i = 0 ; i < sc->neqs; i++) {
2354 aic = &sc->aic_obj[i];
2355 /* When setting the static eq delay from the user space */
2364 rxpkts = rq->rx_stats.rx_pkts;
2366 tx_reqs = wq->tx_stats.tx_reqs;
2369 if (!aic->ticks || now < aic->ticks ||
2370 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2371 aic->prev_rxpkts = rxpkts;
2372 aic->prev_txreqs = tx_reqs;
2377 delta = ticks_to_msecs(now - aic->ticks);
2379 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2380 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2381 eqd = (pps / 15000) << 2;
2385 /* Make sure that the eq delay is in the known range */
2386 eqd = min(eqd, aic->max_eqd);
2387 eqd = max(eqd, aic->min_eqd);
2389 aic->prev_rxpkts = rxpkts;
2390 aic->prev_txreqs = tx_reqs;
2394 if (eqd != aic->cur_eqd) {
2395 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2396 set_eqd[num].eq_id = eqo->eq_id;
2402 /* Is there atleast one eq that needs to be modified? */
2403 for(i = 0; i < num; i += 8) {
2405 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2407 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2412 static void oce_detect_hw_error(POCE_SOFTC sc)
2415 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2416 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2423 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2424 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2425 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2426 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2429 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2430 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2431 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2432 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2434 ue_low = (ue_low & ~ue_low_mask);
2435 ue_high = (ue_high & ~ue_high_mask);
2438 /* On certain platforms BE hardware can indicate spurious UEs.
2439 * Allow the h/w to stop working completely in case of a real UE.
2440 * Hence not setting the hw_error for UE detection.
2442 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2443 sc->hw_error = TRUE;
2444 device_printf(sc->dev, "Error detected in the card\n");
2447 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2448 device_printf(sc->dev,
2449 "ERR: sliport status 0x%x\n", sliport_status);
2450 device_printf(sc->dev,
2451 "ERR: sliport error1 0x%x\n", sliport_err1);
2452 device_printf(sc->dev,
2453 "ERR: sliport error2 0x%x\n", sliport_err2);
2457 for (i = 0; ue_low; ue_low >>= 1, i++) {
2459 device_printf(sc->dev, "UE: %s bit set\n",
2460 ue_status_low_desc[i]);
2465 for (i = 0; ue_high; ue_high >>= 1, i++) {
2467 device_printf(sc->dev, "UE: %s bit set\n",
2468 ue_status_hi_desc[i]);
2476 oce_local_timer(void *arg)
2478 POCE_SOFTC sc = arg;
2481 oce_detect_hw_error(sc);
2482 oce_refresh_nic_stats(sc);
2483 oce_refresh_queue_stats(sc);
2484 oce_mac_addr_set(sc);
2487 for (i = 0; i < sc->nwqs; i++)
2488 oce_tx_restart(sc, sc->wq[i]);
2490 /* calculate and set the eq delay for optimal interrupt rate */
2491 if (IS_BE(sc) || IS_SH(sc))
2492 oce_eqd_set_periodic(sc);
2494 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2498 oce_tx_compl_clean(POCE_SOFTC sc)
2501 int i = 0, timeo = 0, num_wqes = 0;
2502 int pending_txqs = sc->nwqs;
2504 /* Stop polling for compls when HW has been silent for 10ms or
2505 * hw_error or no outstanding completions expected
2508 pending_txqs = sc->nwqs;
2510 for_all_wq_queues(sc, wq, i) {
2511 num_wqes = oce_wq_handler(wq);
2516 if(!wq->ring->num_used)
2520 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2526 for_all_wq_queues(sc, wq, i) {
2527 while(wq->ring->num_used) {
2528 LOCK(&wq->tx_compl_lock);
2529 oce_process_tx_completion(wq);
2530 UNLOCK(&wq->tx_compl_lock);
2536 /* NOTE : This should only be called holding
2540 oce_if_deactivate(POCE_SOFTC sc)
2547 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2549 oce_tx_compl_clean(sc);
2551 /* Stop intrs and finish any bottom halves pending */
2552 oce_hw_intr_disable(sc);
2554 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2555 any other lock. So unlock device lock and require after
2556 completing taskqueue_drain.
2558 UNLOCK(&sc->dev_lock);
2559 for (i = 0; i < sc->intr_count; i++) {
2560 if (sc->intrs[i].tq != NULL) {
2561 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2564 LOCK(&sc->dev_lock);
2566 /* Delete RX queue in card with flush param */
2569 /* Invalidate any pending cq and eq entries*/
2570 for_all_evnt_queues(sc, eq, i)
2572 for_all_rq_queues(sc, rq, i)
2573 oce_drain_rq_cq(rq);
2574 for_all_wq_queues(sc, wq, i)
2575 oce_drain_wq_cq(wq);
2577 /* But still we need to get MCC aync events.
2578 So enable intrs and also arm first EQ
2580 oce_hw_intr_enable(sc);
2581 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2588 oce_if_activate(POCE_SOFTC sc)
2595 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2597 oce_hw_intr_disable(sc);
2601 for_all_rq_queues(sc, rq, i) {
2602 rc = oce_start_rq(rq);
2604 device_printf(sc->dev, "Unable to start RX\n");
2607 for_all_wq_queues(sc, wq, i) {
2608 rc = oce_start_wq(wq);
2610 device_printf(sc->dev, "Unable to start TX\n");
2614 for_all_evnt_queues(sc, eq, i)
2615 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2617 oce_hw_intr_enable(sc);
2622 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2624 /* Update Link status */
2625 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2626 ASYNC_EVENT_LINK_UP) {
2627 sc->link_status = ASYNC_EVENT_LINK_UP;
2628 if_link_state_change(sc->ifp, LINK_STATE_UP);
2630 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2631 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2636 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2637 struct oce_async_evt_grp5_os2bmc *evt)
2639 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2640 if (evt->u.s.mgmt_enable)
2641 sc->flags |= OCE_FLAGS_OS2BMC;
2645 sc->bmc_filt_mask = evt->u.s.arp_filter;
2646 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2647 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2648 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2649 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2650 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2651 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2652 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2653 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2657 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2659 struct oce_async_event_grp5_pvid_state *gcqe;
2660 struct oce_async_evt_grp5_os2bmc *bmccqe;
2662 switch (cqe->u0.s.async_type) {
2663 case ASYNC_EVENT_PVID_STATE:
2665 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2667 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2671 case ASYNC_EVENT_OS2BMC:
2672 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2673 oce_async_grp5_osbmc_process(sc, bmccqe);
2680 /* Handle the Completion Queue for the Mailbox/Async notifications */
2682 oce_mq_handler(void *arg)
2684 struct oce_mq *mq = (struct oce_mq *)arg;
2685 POCE_SOFTC sc = mq->parent;
2686 struct oce_cq *cq = mq->cq;
2687 int num_cqes = 0, evt_type = 0, optype = 0;
2688 struct oce_mq_cqe *cqe;
2689 struct oce_async_cqe_link_state *acqe;
2690 struct oce_async_event_qnq *dbgcqe;
2693 bus_dmamap_sync(cq->ring->dma.tag,
2694 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2695 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2697 while (cqe->u0.dw[3]) {
2698 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2699 if (cqe->u0.s.async_event) {
2700 evt_type = cqe->u0.s.event_type;
2701 optype = cqe->u0.s.async_type;
2702 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2703 /* Link status evt */
2704 acqe = (struct oce_async_cqe_link_state *)cqe;
2705 process_link_state(sc, acqe);
2706 } else if (evt_type == ASYNC_EVENT_GRP5) {
2707 oce_process_grp5_events(sc, cqe);
2708 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2709 optype == ASYNC_EVENT_DEBUG_QNQ) {
2710 dbgcqe = (struct oce_async_event_qnq *)cqe;
2712 sc->qnqid = dbgcqe->vlan_tag;
2713 sc->qnq_debug_event = TRUE;
2717 RING_GET(cq->ring, 1);
2718 bus_dmamap_sync(cq->ring->dma.tag,
2719 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2720 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2725 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2732 setup_max_queues_want(POCE_SOFTC sc)
2734 /* Check if it is FLEX machine. Is so dont use RSS */
2735 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2736 (sc->function_mode & FNM_UMC_MODE) ||
2737 (sc->function_mode & FNM_VNIC_MODE) ||
2738 (!is_rss_enabled(sc)) ||
2743 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2744 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2747 if (IS_BE2(sc) && is_rss_enabled(sc))
2748 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2753 update_queues_got(POCE_SOFTC sc)
2755 if (is_rss_enabled(sc)) {
2756 sc->nrqs = sc->intr_count + 1;
2757 sc->nwqs = sc->intr_count;
2768 oce_check_ipv6_ext_hdr(struct mbuf *m)
2770 struct ether_header *eh = mtod(m, struct ether_header *);
2771 caddr_t m_datatemp = m->m_data;
2773 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2774 m->m_data += sizeof(struct ether_header);
2775 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2777 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2778 (ip6->ip6_nxt != IPPROTO_UDP)){
2779 struct ip6_ext *ip6e = NULL;
2780 m->m_data += sizeof(struct ip6_hdr);
2782 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2783 if(ip6e->ip6e_len == 0xff) {
2784 m->m_data = m_datatemp;
2788 m->m_data = m_datatemp;
2794 is_be3_a1(POCE_SOFTC sc)
2796 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2802 static struct mbuf *
2803 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2805 uint16_t vlan_tag = 0;
2810 /* Embed vlan tag in the packet if it is not part of it */
2811 if(m->m_flags & M_VLANTAG) {
2812 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2813 m->m_flags &= ~M_VLANTAG;
2816 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2819 vlan_tag = sc->pvid;
2825 m = ether_vlanencap(m, vlan_tag);
2829 m = ether_vlanencap(m, sc->qnqid);
2838 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2840 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2841 oce_check_ipv6_ext_hdr(m)) {
2848 oce_get_config(POCE_SOFTC sc)
2851 uint32_t max_rss = 0;
2853 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2854 max_rss = OCE_LEGACY_MODE_RSS;
2856 max_rss = OCE_MAX_RSS;
2859 rc = oce_get_profile_config(sc, max_rss);
2861 sc->nwqs = OCE_MAX_WQ;
2862 sc->nrssqs = max_rss;
2863 sc->nrqs = sc->nrssqs + 1;
2866 else { /* For BE3 don't rely on fw for determining the resources */
2867 sc->nrssqs = max_rss;
2868 sc->nrqs = sc->nrssqs + 1;
2869 sc->nwqs = OCE_MAX_WQ;
2870 sc->max_vlans = MAX_VLANFILTER_SIZE;
2875 oce_rdma_close(void)
2877 if (oce_rdma_if != NULL) {
2883 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2885 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2889 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2892 struct oce_dev_info di;
2895 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2899 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2900 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2904 rdma_info->close = oce_rdma_close;
2905 rdma_info->mbox_post = oce_mbox_post;
2906 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2907 rdma_info->get_mac_addr = oce_get_mac_addr;
2909 oce_rdma_if = rdma_if;
2912 while (sc != NULL) {
2913 if (oce_rdma_if->announce != NULL) {
2914 memset(&di, 0, sizeof(di));
2918 di.db_bhandle = sc->db_bhandle;
2919 di.db_btag = sc->db_btag;
2920 di.db_page_size = 4096;
2921 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2922 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2923 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2924 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2926 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2928 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2929 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2930 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2931 di.msix.start_vector = sc->intr_count;
2932 for (i=0; i<di.msix.num_vectors; i++) {
2933 di.msix.vector_list[i] = sc->intrs[i].vector;
2937 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2938 di.vendor_id = pci_get_vendor(sc->dev);
2939 di.dev_id = pci_get_device(sc->dev);
2941 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2942 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2945 rdma_if->announce(&di);
2954 oce_read_env_variables( POCE_SOFTC sc )
2959 /* read if user wants to enable hwlro or swlro */
2960 //value = getenv("oce_enable_hwlro");
2961 if(value && IS_SH(sc)) {
2962 sc->enable_hwlro = strtol(value, NULL, 10);
2963 if(sc->enable_hwlro) {
2964 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
2966 device_printf(sc->dev, "no hardware lro support\n");
2967 device_printf(sc->dev, "software lro enabled\n");
2968 sc->enable_hwlro = 0;
2970 device_printf(sc->dev, "hardware lro enabled\n");
2971 oce_max_rsp_handled = 32;
2974 device_printf(sc->dev, "software lro enabled\n");
2977 sc->enable_hwlro = 0;
2980 /* read mbuf size */
2981 //value = getenv("oce_rq_buf_size");
2982 if(value && IS_SH(sc)) {
2983 oce_rq_buf_size = strtol(value, NULL, 10);
2984 switch(oce_rq_buf_size) {
2992 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
2993 oce_rq_buf_size = 2048;