2 * Copyright (C) 2013 Emulex
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are met:
8 * 1. Redistributions of source code must retain the above copyright notice,
9 * this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the Emulex Corporation nor the names of its
16 * contributors may be used to endorse or promote products derived from
17 * this software without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
31 * Contact Information:
32 * freebsd-drivers@emulex.com
36 * Costa Mesa, CA 92626
41 #include "opt_inet6.h"
47 #define is_tso_pkt(m) (m->m_pkthdr.csum_flags & CSUM_TSO)
49 /* UE Status Low CSR */
50 static char *ue_status_low_desc[] = {
85 /* UE Status High CSR */
86 static char *ue_status_hi_desc[] = {
121 struct oce_common_cqe_info{
123 uint8_t l4_cksum_pass:1;
124 uint8_t ip_cksum_pass:1;
125 uint8_t ipv6_frame:1;
134 /* Driver entry points prototypes */
135 static int oce_probe(device_t dev);
136 static int oce_attach(device_t dev);
137 static int oce_detach(device_t dev);
138 static int oce_shutdown(device_t dev);
139 static int oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
140 static void oce_init(void *xsc);
141 static int oce_multiq_start(struct ifnet *ifp, struct mbuf *m);
142 static void oce_multiq_flush(struct ifnet *ifp);
144 /* Driver interrupt routines protypes */
145 static void oce_intr(void *arg, int pending);
146 static int oce_setup_intr(POCE_SOFTC sc);
147 static int oce_fast_isr(void *arg);
148 static int oce_alloc_intr(POCE_SOFTC sc, int vector,
149 void (*isr) (void *arg, int pending));
151 /* Media callbacks prototypes */
152 static void oce_media_status(struct ifnet *ifp, struct ifmediareq *req);
153 static int oce_media_change(struct ifnet *ifp);
155 /* Transmit routines prototypes */
156 static int oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index);
157 static void oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq);
158 static void oce_process_tx_completion(struct oce_wq *wq);
159 static int oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m,
162 /* Receive routines prototypes */
163 static int oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
164 static int oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
165 static void oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe);
166 static void oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq);
167 static uint16_t oce_rq_handler_lro(void *arg);
168 static void oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2);
169 static void oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2);
170 static void oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m);
172 /* Helper function prototypes in this file */
173 static int oce_attach_ifp(POCE_SOFTC sc);
174 static void oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
175 static void oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
176 static int oce_vid_config(POCE_SOFTC sc);
177 static void oce_mac_addr_set(POCE_SOFTC sc);
178 static int oce_handle_passthrough(struct ifnet *ifp, caddr_t data);
179 static void oce_local_timer(void *arg);
180 static void oce_if_deactivate(POCE_SOFTC sc);
181 static void oce_if_activate(POCE_SOFTC sc);
182 static void setup_max_queues_want(POCE_SOFTC sc);
183 static void update_queues_got(POCE_SOFTC sc);
184 static void process_link_state(POCE_SOFTC sc,
185 struct oce_async_cqe_link_state *acqe);
186 static int oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m);
187 static void oce_get_config(POCE_SOFTC sc);
188 static struct mbuf *oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete);
189 static void oce_read_env_variables(POCE_SOFTC sc);
193 #if defined(INET6) || defined(INET)
194 static int oce_init_lro(POCE_SOFTC sc);
195 static struct mbuf * oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp);
198 static device_method_t oce_dispatch[] = {
199 DEVMETHOD(device_probe, oce_probe),
200 DEVMETHOD(device_attach, oce_attach),
201 DEVMETHOD(device_detach, oce_detach),
202 DEVMETHOD(device_shutdown, oce_shutdown),
207 static driver_t oce_driver = {
212 static devclass_t oce_devclass;
215 DRIVER_MODULE(oce, pci, oce_driver, oce_devclass, 0, 0);
216 MODULE_DEPEND(oce, pci, 1, 1, 1);
217 MODULE_DEPEND(oce, ether, 1, 1, 1);
218 MODULE_VERSION(oce, 1);
222 const char component_revision[32] = {"///" COMPONENT_REVISION "///"};
224 /* Module capabilites and parameters */
225 uint32_t oce_max_rsp_handled = OCE_MAX_RSP_HANDLED;
226 uint32_t oce_enable_rss = OCE_MODCAP_RSS;
227 uint32_t oce_rq_buf_size = 2048;
229 TUNABLE_INT("hw.oce.max_rsp_handled", &oce_max_rsp_handled);
230 TUNABLE_INT("hw.oce.enable_rss", &oce_enable_rss);
233 /* Supported devices table */
234 static uint32_t supportedDevices[] = {
235 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE2,
236 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE3,
237 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_BE3,
238 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201,
239 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201_VF,
240 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_SH
243 POCE_SOFTC softc_head = NULL;
244 POCE_SOFTC softc_tail = NULL;
246 struct oce_rdma_if *oce_rdma_if = NULL;
248 /*****************************************************************************
249 * Driver entry points functions *
250 *****************************************************************************/
253 oce_probe(device_t dev)
261 sc = device_get_softc(dev);
262 bzero(sc, sizeof(OCE_SOFTC));
265 vendor = pci_get_vendor(dev);
266 device = pci_get_device(dev);
268 for (i = 0; i < (sizeof(supportedDevices) / sizeof(uint32_t)); i++) {
269 if (vendor == ((supportedDevices[i] >> 16) & 0xffff)) {
270 if (device == (supportedDevices[i] & 0xffff)) {
271 sprintf(str, "%s:%s", "Emulex CNA NIC function",
273 device_set_desc_copy(dev, str);
276 case PCI_PRODUCT_BE2:
277 sc->flags |= OCE_FLAGS_BE2;
279 case PCI_PRODUCT_BE3:
280 sc->flags |= OCE_FLAGS_BE3;
282 case PCI_PRODUCT_XE201:
283 case PCI_PRODUCT_XE201_VF:
284 sc->flags |= OCE_FLAGS_XE201;
287 sc->flags |= OCE_FLAGS_SH;
292 return BUS_PROBE_DEFAULT;
302 oce_attach(device_t dev)
307 sc = device_get_softc(dev);
309 rc = oce_hw_pci_alloc(sc);
313 sc->tx_ring_size = OCE_TX_RING_SIZE;
314 sc->rx_ring_size = OCE_RX_RING_SIZE;
315 /* receive fragment size should be multiple of 2K */
316 sc->rq_frag_size = ((oce_rq_buf_size / 2048) * 2048);
317 sc->flow_control = OCE_DEFAULT_FLOW_CONTROL;
318 sc->promisc = OCE_DEFAULT_PROMISCUOUS;
320 LOCK_CREATE(&sc->bmbx_lock, "Mailbox_lock");
321 LOCK_CREATE(&sc->dev_lock, "Device_lock");
323 /* initialise the hardware */
324 rc = oce_hw_init(sc);
328 oce_read_env_variables(sc);
332 setup_max_queues_want(sc);
334 rc = oce_setup_intr(sc);
338 rc = oce_queue_init_all(sc);
342 rc = oce_attach_ifp(sc);
346 #if defined(INET6) || defined(INET)
347 rc = oce_init_lro(sc);
352 rc = oce_hw_start(sc);
356 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
357 oce_add_vlan, sc, EVENTHANDLER_PRI_FIRST);
358 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
359 oce_del_vlan, sc, EVENTHANDLER_PRI_FIRST);
361 rc = oce_stats_init(sc);
367 callout_init(&sc->timer, CALLOUT_MPSAFE);
368 rc = callout_reset(&sc->timer, 2 * hz, oce_local_timer, sc);
373 if (softc_tail != NULL) {
374 softc_tail->next = sc;
383 callout_drain(&sc->timer);
387 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
389 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
390 oce_hw_intr_disable(sc);
392 #if defined(INET6) || defined(INET)
396 ether_ifdetach(sc->ifp);
399 oce_queue_release_all(sc);
403 oce_dma_free(sc, &sc->bsmbx);
406 LOCK_DESTROY(&sc->dev_lock);
407 LOCK_DESTROY(&sc->bmbx_lock);
414 oce_detach(device_t dev)
416 POCE_SOFTC sc = device_get_softc(dev);
417 POCE_SOFTC poce_sc_tmp, *ppoce_sc_tmp1, poce_sc_tmp2 = NULL;
419 poce_sc_tmp = softc_head;
420 ppoce_sc_tmp1 = &softc_head;
421 while (poce_sc_tmp != NULL) {
422 if (poce_sc_tmp == sc) {
423 *ppoce_sc_tmp1 = sc->next;
424 if (sc->next == NULL) {
425 softc_tail = poce_sc_tmp2;
429 poce_sc_tmp2 = poce_sc_tmp;
430 ppoce_sc_tmp1 = &poce_sc_tmp->next;
431 poce_sc_tmp = poce_sc_tmp->next;
435 oce_if_deactivate(sc);
436 UNLOCK(&sc->dev_lock);
438 callout_drain(&sc->timer);
440 if (sc->vlan_attach != NULL)
441 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
442 if (sc->vlan_detach != NULL)
443 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
445 ether_ifdetach(sc->ifp);
451 bus_generic_detach(dev);
458 oce_shutdown(device_t dev)
462 rc = oce_detach(dev);
469 oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
471 struct ifreq *ifr = (struct ifreq *)data;
472 POCE_SOFTC sc = ifp->if_softc;
481 rc = ifmedia_ioctl(ifp, ifr, &sc->media, command);
485 if (ifr->ifr_mtu > OCE_MAX_MTU)
488 ifp->if_mtu = ifr->ifr_mtu;
492 if (ifp->if_flags & IFF_UP) {
493 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
494 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
497 device_printf(sc->dev, "Interface Up\n");
501 sc->ifp->if_drv_flags &=
502 ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
503 oce_if_deactivate(sc);
505 UNLOCK(&sc->dev_lock);
507 device_printf(sc->dev, "Interface Down\n");
510 if ((ifp->if_flags & IFF_PROMISC) && !sc->promisc) {
511 if (!oce_rxf_set_promiscuous(sc, (1 | (1 << 1))))
513 } else if (!(ifp->if_flags & IFF_PROMISC) && sc->promisc) {
514 if (!oce_rxf_set_promiscuous(sc, 0))
522 rc = oce_hw_update_multicast(sc);
524 device_printf(sc->dev,
525 "Update multicast address failed\n");
529 u = ifr->ifr_reqcap ^ ifp->if_capenable;
531 if (u & IFCAP_TXCSUM) {
532 ifp->if_capenable ^= IFCAP_TXCSUM;
533 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
535 if (IFCAP_TSO & ifp->if_capenable &&
536 !(IFCAP_TXCSUM & ifp->if_capenable)) {
537 ifp->if_capenable &= ~IFCAP_TSO;
538 ifp->if_hwassist &= ~CSUM_TSO;
540 "TSO disabled due to -txcsum.\n");
544 if (u & IFCAP_RXCSUM)
545 ifp->if_capenable ^= IFCAP_RXCSUM;
547 if (u & IFCAP_TSO4) {
548 ifp->if_capenable ^= IFCAP_TSO4;
550 if (IFCAP_TSO & ifp->if_capenable) {
551 if (IFCAP_TXCSUM & ifp->if_capenable)
552 ifp->if_hwassist |= CSUM_TSO;
554 ifp->if_capenable &= ~IFCAP_TSO;
555 ifp->if_hwassist &= ~CSUM_TSO;
557 "Enable txcsum first.\n");
561 ifp->if_hwassist &= ~CSUM_TSO;
564 if (u & IFCAP_VLAN_HWTAGGING)
565 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
567 if (u & IFCAP_VLAN_HWFILTER) {
568 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
571 #if defined(INET6) || defined(INET)
573 ifp->if_capenable ^= IFCAP_LRO;
574 if(sc->enable_hwlro) {
575 if(ifp->if_capenable & IFCAP_LRO) {
576 rc = oce_mbox_nic_set_iface_lro_config(sc, 1);
578 rc = oce_mbox_nic_set_iface_lro_config(sc, 0);
587 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
591 if (i2c.dev_addr != PAGE_NUM_A0 &&
592 i2c.dev_addr != PAGE_NUM_A2) {
597 if (i2c.len > sizeof(i2c.data)) {
602 rc = oce_mbox_read_transrecv_data(sc, i2c.dev_addr);
608 if (i2c.dev_addr == PAGE_NUM_A0)
611 offset = TRANSCEIVER_A0_SIZE + i2c.offset;
613 memcpy(&i2c.data[0], &sfp_vpd_dump_buffer[offset], i2c.len);
615 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
619 rc = oce_handle_passthrough(ifp, data);
622 rc = ether_ioctl(ifp, command, data);
637 if (sc->ifp->if_flags & IFF_UP) {
638 oce_if_deactivate(sc);
642 UNLOCK(&sc->dev_lock);
648 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
650 POCE_SOFTC sc = ifp->if_softc;
651 struct oce_wq *wq = NULL;
655 if (!sc->link_status)
658 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
659 queue_index = m->m_pkthdr.flowid % sc->nwqs;
661 wq = sc->wq[queue_index];
664 status = oce_multiq_transmit(ifp, m, wq);
665 UNLOCK(&wq->tx_lock);
673 oce_multiq_flush(struct ifnet *ifp)
675 POCE_SOFTC sc = ifp->if_softc;
679 for (i = 0; i < sc->nwqs; i++) {
680 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
688 /*****************************************************************************
689 * Driver interrupt routines functions *
690 *****************************************************************************/
693 oce_intr(void *arg, int pending)
696 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
697 POCE_SOFTC sc = ii->sc;
698 struct oce_eq *eq = ii->eq;
700 struct oce_cq *cq = NULL;
704 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
705 BUS_DMASYNC_POSTWRITE);
707 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
711 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
712 BUS_DMASYNC_POSTWRITE);
713 RING_GET(eq->ring, 1);
719 goto eq_arm; /* Spurious */
721 /* Clear EQ entries, but dont arm */
722 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
724 /* Process TX, RX and MCC. But dont arm CQ*/
725 for (i = 0; i < eq->cq_valid; i++) {
727 (*cq->cq_handler)(cq->cb_arg);
730 /* Arm all cqs connected to this EQ */
731 for (i = 0; i < eq->cq_valid; i++) {
733 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
737 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
744 oce_setup_intr(POCE_SOFTC sc)
746 int rc = 0, use_intx = 0;
747 int vector = 0, req_vectors = 0;
748 int tot_req_vectors, tot_vectors;
750 if (is_rss_enabled(sc))
751 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
755 tot_req_vectors = req_vectors;
756 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
757 if (req_vectors > 1) {
758 tot_req_vectors += OCE_RDMA_VECTORS;
759 sc->roce_intr_count = OCE_RDMA_VECTORS;
763 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
764 sc->intr_count = req_vectors;
765 tot_vectors = tot_req_vectors;
766 rc = pci_alloc_msix(sc->dev, &tot_vectors);
769 pci_release_msi(sc->dev);
771 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
772 if (tot_vectors < tot_req_vectors) {
773 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
774 sc->roce_intr_count = (tot_vectors / 2);
776 sc->intr_count = tot_vectors - sc->roce_intr_count;
779 sc->intr_count = tot_vectors;
781 sc->flags |= OCE_FLAGS_USING_MSIX;
789 /* Scale number of queues based on intr we got */
790 update_queues_got(sc);
793 device_printf(sc->dev, "Using legacy interrupt\n");
794 rc = oce_alloc_intr(sc, vector, oce_intr);
798 for (; vector < sc->intr_count; vector++) {
799 rc = oce_alloc_intr(sc, vector, oce_intr);
813 oce_fast_isr(void *arg)
815 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
816 POCE_SOFTC sc = ii->sc;
821 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
823 taskqueue_enqueue(ii->tq, &ii->task);
827 return FILTER_HANDLED;
832 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
834 POCE_INTR_INFO ii = &sc->intrs[vector];
837 if (vector >= OCE_MAX_EQ)
840 /* Set the resource id for the interrupt.
841 * MSIx is vector + 1 for the resource id,
842 * INTx is 0 for the resource id.
844 if (sc->flags & OCE_FLAGS_USING_MSIX)
848 ii->intr_res = bus_alloc_resource_any(sc->dev,
850 &rr, RF_ACTIVE|RF_SHAREABLE);
852 if (ii->intr_res == NULL) {
853 device_printf(sc->dev,
854 "Could not allocate interrupt\n");
859 TASK_INIT(&ii->task, 0, isr, ii);
861 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
862 ii->tq = taskqueue_create_fast(ii->task_name,
864 taskqueue_thread_enqueue,
866 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
867 device_get_nameunit(sc->dev));
870 rc = bus_setup_intr(sc->dev,
873 oce_fast_isr, NULL, ii, &ii->tag);
880 oce_intr_free(POCE_SOFTC sc)
884 for (i = 0; i < sc->intr_count; i++) {
886 if (sc->intrs[i].tag != NULL)
887 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
889 if (sc->intrs[i].tq != NULL)
890 taskqueue_free(sc->intrs[i].tq);
892 if (sc->intrs[i].intr_res != NULL)
893 bus_release_resource(sc->dev, SYS_RES_IRQ,
895 sc->intrs[i].intr_res);
896 sc->intrs[i].tag = NULL;
897 sc->intrs[i].intr_res = NULL;
900 if (sc->flags & OCE_FLAGS_USING_MSIX)
901 pci_release_msi(sc->dev);
907 /******************************************************************************
908 * Media callbacks functions *
909 ******************************************************************************/
912 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
914 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
917 req->ifm_status = IFM_AVALID;
918 req->ifm_active = IFM_ETHER;
920 if (sc->link_status == 1)
921 req->ifm_status |= IFM_ACTIVE;
925 switch (sc->link_speed) {
926 case 1: /* 10 Mbps */
927 req->ifm_active |= IFM_10_T | IFM_FDX;
930 case 2: /* 100 Mbps */
931 req->ifm_active |= IFM_100_TX | IFM_FDX;
935 req->ifm_active |= IFM_1000_T | IFM_FDX;
938 case 4: /* 10 Gbps */
939 req->ifm_active |= IFM_10G_SR | IFM_FDX;
942 case 5: /* 20 Gbps */
943 req->ifm_active |= IFM_10G_SR | IFM_FDX;
946 case 6: /* 25 Gbps */
947 req->ifm_active |= IFM_10G_SR | IFM_FDX;
950 case 7: /* 40 Gbps */
951 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
964 oce_media_change(struct ifnet *ifp)
970 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
971 struct mbuf *m, boolean_t *os2bmc,
974 struct ether_header *eh = NULL;
976 eh = mtod(m, struct ether_header *);
978 if (!is_os2bmc_enabled(sc) || *os2bmc) {
982 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
985 if (is_mc_allowed_on_bmc(sc, eh) ||
986 is_bc_allowed_on_bmc(sc, eh) ||
987 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
992 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
993 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
994 uint8_t nexthdr = ip6->ip6_nxt;
995 if (nexthdr == IPPROTO_ICMPV6) {
996 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
997 switch (icmp6->icmp6_type) {
998 case ND_ROUTER_ADVERT:
999 *os2bmc = is_ipv6_ra_filt_enabled(sc);
1001 case ND_NEIGHBOR_ADVERT:
1002 *os2bmc = is_ipv6_na_filt_enabled(sc);
1010 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
1011 struct ip *ip = mtod(m, struct ip *);
1012 int iphlen = ip->ip_hl << 2;
1013 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
1014 switch (uh->uh_dport) {
1015 case DHCP_CLIENT_PORT:
1016 *os2bmc = is_dhcp_client_filt_enabled(sc);
1018 case DHCP_SERVER_PORT:
1019 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
1021 case NET_BIOS_PORT1:
1022 case NET_BIOS_PORT2:
1023 *os2bmc = is_nbios_filt_enabled(sc);
1025 case DHCPV6_RAS_PORT:
1026 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1034 *m_new = m_dup(m, M_NOWAIT);
1039 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1045 /*****************************************************************************
1046 * Transmit routines functions *
1047 *****************************************************************************/
1050 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1052 int rc = 0, i, retry_cnt = 0;
1053 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1054 struct mbuf *m, *m_temp, *m_new = NULL;
1055 struct oce_wq *wq = sc->wq[wq_index];
1056 struct oce_packet_desc *pd;
1057 struct oce_nic_hdr_wqe *nichdr;
1058 struct oce_nic_frag_wqe *nicfrag;
1059 struct ether_header *eh = NULL;
1062 boolean_t complete = TRUE;
1063 boolean_t os2bmc = FALSE;
1069 if (!(m->m_flags & M_PKTHDR)) {
1074 /* Don't allow non-TSO packets longer than MTU */
1075 if (!is_tso_pkt(m)) {
1076 eh = mtod(m, struct ether_header *);
1077 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1081 if(oce_tx_asic_stall_verify(sc, m)) {
1082 m = oce_insert_vlan_tag(sc, m, &complete);
1084 device_printf(sc->dev, "Insertion unsuccessful\n");
1090 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1091 * may cause a transmit stall on that port. So the work-around is to
1092 * pad short packets (<= 32 bytes) to a 36-byte length.
1094 if(IS_SH(sc) || IS_XE201(sc) ) {
1095 if(m->m_pkthdr.len <= 32) {
1097 bzero((void *)buf, 36);
1098 m_append(m, (36 - m->m_pkthdr.len), buf);
1103 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1104 /* consolidate packet buffers for TSO/LSO segment offload */
1105 #if defined(INET6) || defined(INET)
1106 m = oce_tso_setup(sc, mpp);
1117 pd = &wq->pckts[wq->pkt_desc_head];
1120 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1122 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1124 num_wqes = pd->nsegs + 1;
1125 if (IS_BE(sc) || IS_SH(sc)) {
1126 /*Dummy required only for BE3.*/
1130 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1131 bus_dmamap_unload(wq->tag, pd->map);
1134 atomic_store_rel_int(&wq->pkt_desc_head,
1135 (wq->pkt_desc_head + 1) % \
1136 OCE_WQ_PACKET_ARRAY_SIZE);
1137 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1141 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1142 nichdr->u0.dw[0] = 0;
1143 nichdr->u0.dw[1] = 0;
1144 nichdr->u0.dw[2] = 0;
1145 nichdr->u0.dw[3] = 0;
1147 nichdr->u0.s.complete = complete;
1148 nichdr->u0.s.mgmt = os2bmc;
1149 nichdr->u0.s.event = 1;
1150 nichdr->u0.s.crc = 1;
1151 nichdr->u0.s.forward = 0;
1152 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1153 nichdr->u0.s.udpcs =
1154 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1155 nichdr->u0.s.tcpcs =
1156 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1157 nichdr->u0.s.num_wqe = num_wqes;
1158 nichdr->u0.s.total_length = m->m_pkthdr.len;
1160 if (m->m_flags & M_VLANTAG) {
1161 nichdr->u0.s.vlan = 1; /*Vlan present*/
1162 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1165 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1166 if (m->m_pkthdr.tso_segsz) {
1167 nichdr->u0.s.lso = 1;
1168 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1170 if (!IS_BE(sc) || !IS_SH(sc))
1171 nichdr->u0.s.ipcs = 1;
1174 RING_PUT(wq->ring, 1);
1175 atomic_add_int(&wq->ring->num_used, 1);
1177 for (i = 0; i < pd->nsegs; i++) {
1179 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1180 struct oce_nic_frag_wqe);
1181 nicfrag->u0.s.rsvd0 = 0;
1182 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1183 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1184 nicfrag->u0.s.frag_len = segs[i].ds_len;
1185 pd->wqe_idx = wq->ring->pidx;
1186 RING_PUT(wq->ring, 1);
1187 atomic_add_int(&wq->ring->num_used, 1);
1189 if (num_wqes > (pd->nsegs + 1)) {
1191 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1192 struct oce_nic_frag_wqe);
1193 nicfrag->u0.dw[0] = 0;
1194 nicfrag->u0.dw[1] = 0;
1195 nicfrag->u0.dw[2] = 0;
1196 nicfrag->u0.dw[3] = 0;
1197 pd->wqe_idx = wq->ring->pidx;
1198 RING_PUT(wq->ring, 1);
1199 atomic_add_int(&wq->ring->num_used, 1);
1203 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1204 wq->tx_stats.tx_reqs++;
1205 wq->tx_stats.tx_wrbs += num_wqes;
1206 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1207 wq->tx_stats.tx_pkts++;
1209 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1210 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1211 reg_value = (num_wqes << 16) | wq->wq_id;
1213 /* if os2bmc is not enabled or if the pkt is already tagged as
1216 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1218 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1220 } else if (rc == EFBIG) {
1221 if (retry_cnt == 0) {
1222 m_temp = m_defrag(m, M_NOWAIT);
1227 retry_cnt = retry_cnt + 1;
1231 } else if (rc == ENOMEM)
1251 oce_process_tx_completion(struct oce_wq *wq)
1253 struct oce_packet_desc *pd;
1254 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1257 pd = &wq->pckts[wq->pkt_desc_tail];
1258 atomic_store_rel_int(&wq->pkt_desc_tail,
1259 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1260 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1261 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1262 bus_dmamap_unload(wq->tag, pd->map);
1269 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1270 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1271 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1272 oce_tx_restart(sc, wq);
1279 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1282 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1285 #if __FreeBSD_version >= 800000
1286 if (!drbr_empty(sc->ifp, wq->br))
1288 if (!IFQ_DRV_IS_EMPTY(&sc->ifp->if_snd))
1290 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1295 #if defined(INET6) || defined(INET)
1296 static struct mbuf *
1297 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1304 struct ip6_hdr *ip6;
1306 struct ether_vlan_header *eh;
1309 int total_len = 0, ehdrlen = 0;
1313 if (M_WRITABLE(m) == 0) {
1314 m = m_dup(*mpp, M_NOWAIT);
1321 eh = mtod(m, struct ether_vlan_header *);
1322 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1323 etype = ntohs(eh->evl_proto);
1324 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1326 etype = ntohs(eh->evl_encap_proto);
1327 ehdrlen = ETHER_HDR_LEN;
1333 ip = (struct ip *)(m->m_data + ehdrlen);
1334 if (ip->ip_p != IPPROTO_TCP)
1336 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1338 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1342 case ETHERTYPE_IPV6:
1343 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1344 if (ip6->ip6_nxt != IPPROTO_TCP)
1346 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1348 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1355 m = m_pullup(m, total_len);
1362 #endif /* INET6 || INET */
1365 oce_tx_task(void *arg, int npending)
1367 struct oce_wq *wq = arg;
1368 POCE_SOFTC sc = wq->parent;
1369 struct ifnet *ifp = sc->ifp;
1372 #if __FreeBSD_version >= 800000
1374 rc = oce_multiq_transmit(ifp, NULL, wq);
1376 device_printf(sc->dev,
1377 "TX[%d] restart failed\n", wq->queue_index);
1379 UNLOCK(&wq->tx_lock);
1388 oce_start(struct ifnet *ifp)
1390 POCE_SOFTC sc = ifp->if_softc;
1393 int def_q = 0; /* Defualt tx queue is 0*/
1395 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1399 if (!sc->link_status)
1403 IF_DEQUEUE(&sc->ifp->if_snd, m);
1407 LOCK(&sc->wq[def_q]->tx_lock);
1408 rc = oce_tx(sc, &m, def_q);
1409 UNLOCK(&sc->wq[def_q]->tx_lock);
1412 sc->wq[def_q]->tx_stats.tx_stops ++;
1413 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1414 IFQ_DRV_PREPEND(&ifp->if_snd, m);
1420 ETHER_BPF_MTAP(ifp, m);
1428 /* Handle the Completion Queue for transmit */
1430 oce_wq_handler(void *arg)
1432 struct oce_wq *wq = (struct oce_wq *)arg;
1433 POCE_SOFTC sc = wq->parent;
1434 struct oce_cq *cq = wq->cq;
1435 struct oce_nic_tx_cqe *cqe;
1438 LOCK(&wq->tx_compl_lock);
1439 bus_dmamap_sync(cq->ring->dma.tag,
1440 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1441 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1442 while (cqe->u0.dw[3]) {
1443 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe));
1445 wq->ring->cidx = cqe->u0.s.wqe_index + 1;
1446 if (wq->ring->cidx >= wq->ring->num_items)
1447 wq->ring->cidx -= wq->ring->num_items;
1449 oce_process_tx_completion(wq);
1450 wq->tx_stats.tx_compl++;
1452 RING_GET(cq->ring, 1);
1453 bus_dmamap_sync(cq->ring->dma.tag,
1454 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1456 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1461 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
1463 UNLOCK(&wq->tx_compl_lock);
1469 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq)
1471 POCE_SOFTC sc = ifp->if_softc;
1472 int status = 0, queue_index = 0;
1473 struct mbuf *next = NULL;
1474 struct buf_ring *br = NULL;
1477 queue_index = wq->queue_index;
1479 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1482 status = drbr_enqueue(ifp, br, m);
1487 if ((status = drbr_enqueue(ifp, br, m)) != 0)
1490 while ((next = drbr_peek(ifp, br)) != NULL) {
1491 if (oce_tx(sc, &next, queue_index)) {
1493 drbr_advance(ifp, br);
1495 drbr_putback(ifp, br, next);
1496 wq->tx_stats.tx_stops ++;
1497 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1501 drbr_advance(ifp, br);
1502 if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
1503 if (next->m_flags & M_MCAST)
1504 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1505 ETHER_BPF_MTAP(ifp, next);
1514 /*****************************************************************************
1515 * Receive routines functions *
1516 *****************************************************************************/
1519 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2)
1522 struct ether_header *eh = NULL;
1523 struct tcphdr *tcp_hdr = NULL;
1524 struct ip *ip4_hdr = NULL;
1525 struct ip6_hdr *ip6 = NULL;
1526 uint32_t payload_len = 0;
1528 eh = mtod(m, struct ether_header *);
1529 /* correct IP header */
1530 if(!cqe2->ipv6_frame) {
1531 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header));
1532 ip4_hdr->ip_ttl = cqe2->frame_lifespan;
1533 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header));
1534 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip));
1536 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header));
1537 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan;
1538 payload_len = cqe2->coalesced_size - sizeof(struct ether_header)
1539 - sizeof(struct ip6_hdr);
1540 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len);
1541 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr));
1544 /* correct tcp header */
1545 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num);
1547 tcp_hdr->th_flags |= TH_PUSH;
1549 tcp_hdr->th_win = htons(cqe2->tcp_window);
1550 tcp_hdr->th_sum = 0xffff;
1552 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2);
1553 *p = cqe1->tcp_timestamp_val;
1554 *(p+1) = cqe1->tcp_timestamp_ecr;
1561 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m)
1563 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1564 uint32_t i = 0, frag_len = 0;
1565 uint32_t len = cqe_info->pkt_size;
1566 struct oce_packet_desc *pd;
1567 struct mbuf *tail = NULL;
1569 for (i = 0; i < cqe_info->num_frags; i++) {
1570 if (rq->ring->cidx == rq->ring->pidx) {
1571 device_printf(sc->dev,
1572 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n");
1575 pd = &rq->pckts[rq->ring->cidx];
1577 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1578 bus_dmamap_unload(rq->tag, pd->map);
1579 RING_GET(rq->ring, 1);
1582 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len;
1583 pd->mbuf->m_len = frag_len;
1586 /* additional fragments */
1587 pd->mbuf->m_flags &= ~M_PKTHDR;
1588 tail->m_next = pd->mbuf;
1590 tail->m_nextpkt = NULL;
1593 /* first fragment, fill out much of the packet header */
1594 pd->mbuf->m_pkthdr.len = len;
1596 pd->mbuf->m_nextpkt = NULL;
1597 pd->mbuf->m_pkthdr.csum_flags = 0;
1598 if (IF_CSUM_ENABLED(sc)) {
1599 if (cqe_info->l4_cksum_pass) {
1600 if(!cqe_info->ipv6_frame) { /* IPV4 */
1601 pd->mbuf->m_pkthdr.csum_flags |=
1602 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1603 }else { /* IPV6 frame */
1605 pd->mbuf->m_pkthdr.csum_flags |=
1606 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1609 pd->mbuf->m_pkthdr.csum_data = 0xffff;
1611 if (cqe_info->ip_cksum_pass) {
1612 pd->mbuf->m_pkthdr.csum_flags |=
1613 (CSUM_IP_CHECKED|CSUM_IP_VALID);
1616 *m = tail = pd->mbuf;
1626 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2)
1628 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1629 struct nic_hwlro_cqe_part1 *cqe1 = NULL;
1630 struct mbuf *m = NULL;
1631 struct oce_common_cqe_info cq_info;
1635 cq_info.pkt_size = cqe->pkt_size;
1636 cq_info.vtag = cqe->vlan_tag;
1637 cq_info.l4_cksum_pass = cqe->l4_cksum_pass;
1638 cq_info.ip_cksum_pass = cqe->ip_cksum_pass;
1639 cq_info.ipv6_frame = cqe->ipv6_frame;
1640 cq_info.vtp = cqe->vtp;
1641 cq_info.qnq = cqe->qnq;
1643 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe;
1644 cq_info.pkt_size = cqe2->coalesced_size;
1645 cq_info.vtag = cqe2->vlan_tag;
1646 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass;
1647 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass;
1648 cq_info.ipv6_frame = cqe2->ipv6_frame;
1649 cq_info.vtp = cqe2->vtp;
1650 cq_info.qnq = cqe1->qnq;
1653 cq_info.vtag = BSWAP_16(cq_info.vtag);
1655 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size;
1656 if(cq_info.pkt_size % rq->cfg.frag_size)
1657 cq_info.num_frags++;
1659 oce_rx_mbuf_chain(rq, &cq_info, &m);
1663 //assert(cqe2->valid != 0);
1665 //assert(cqe2->cqe_type != 2);
1666 oce_correct_header(m, cqe1, cqe2);
1669 m->m_pkthdr.rcvif = sc->ifp;
1670 #if __FreeBSD_version >= 800000
1671 if (rq->queue_index)
1672 m->m_pkthdr.flowid = (rq->queue_index - 1);
1674 m->m_pkthdr.flowid = rq->queue_index;
1675 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1677 /* This deternies if vlan tag is Valid */
1679 if (sc->function_mode & FNM_FLEX10_MODE) {
1680 /* FLEX10. If QnQ is not set, neglect VLAN */
1682 m->m_pkthdr.ether_vtag = cq_info.vtag;
1683 m->m_flags |= M_VLANTAG;
1685 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1686 /* In UMC mode generally pvid will be striped by
1687 hw. But in some cases we have seen it comes
1688 with pvid. So if pvid == vlan, neglect vlan.
1690 m->m_pkthdr.ether_vtag = cq_info.vtag;
1691 m->m_flags |= M_VLANTAG;
1694 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1696 (*sc->ifp->if_input) (sc->ifp, m);
1698 /* Update rx stats per queue */
1699 rq->rx_stats.rx_pkts++;
1700 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1701 rq->rx_stats.rx_frags += cq_info.num_frags;
1702 rq->rx_stats.rx_ucast_pkts++;
1708 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1710 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1712 struct mbuf *m = NULL;
1713 struct oce_common_cqe_info cq_info;
1716 /* Is it a flush compl that has no data */
1717 if(!cqe->u0.s.num_fragments)
1720 len = cqe->u0.s.pkt_size;
1722 /*partial DMA workaround for Lancer*/
1723 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1727 if (!oce_cqe_portid_valid(sc, cqe)) {
1728 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1732 /* Get vlan_tag value */
1733 if(IS_BE(sc) || IS_SH(sc))
1734 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1736 vtag = cqe->u0.s.vlan_tag;
1738 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1739 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1740 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1741 cq_info.num_frags = cqe->u0.s.num_fragments;
1742 cq_info.pkt_size = cqe->u0.s.pkt_size;
1744 oce_rx_mbuf_chain(rq, &cq_info, &m);
1747 m->m_pkthdr.rcvif = sc->ifp;
1748 #if __FreeBSD_version >= 800000
1749 if (rq->queue_index)
1750 m->m_pkthdr.flowid = (rq->queue_index - 1);
1752 m->m_pkthdr.flowid = rq->queue_index;
1753 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1755 /* This deternies if vlan tag is Valid */
1756 if (oce_cqe_vtp_valid(sc, cqe)) {
1757 if (sc->function_mode & FNM_FLEX10_MODE) {
1758 /* FLEX10. If QnQ is not set, neglect VLAN */
1759 if (cqe->u0.s.qnq) {
1760 m->m_pkthdr.ether_vtag = vtag;
1761 m->m_flags |= M_VLANTAG;
1763 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1764 /* In UMC mode generally pvid will be striped by
1765 hw. But in some cases we have seen it comes
1766 with pvid. So if pvid == vlan, neglect vlan.
1768 m->m_pkthdr.ether_vtag = vtag;
1769 m->m_flags |= M_VLANTAG;
1773 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1774 #if defined(INET6) || defined(INET)
1775 /* Try to queue to LRO */
1776 if (IF_LRO_ENABLED(sc) &&
1777 (cqe->u0.s.ip_cksum_pass) &&
1778 (cqe->u0.s.l4_cksum_pass) &&
1779 (!cqe->u0.s.ip_ver) &&
1780 (rq->lro.lro_cnt != 0)) {
1782 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1783 rq->lro_pkts_queued ++;
1786 /* If LRO posting fails then try to post to STACK */
1790 (*sc->ifp->if_input) (sc->ifp, m);
1791 #if defined(INET6) || defined(INET)
1794 /* Update rx stats per queue */
1795 rq->rx_stats.rx_pkts++;
1796 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1797 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1798 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1799 rq->rx_stats.rx_mcast_pkts++;
1800 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1801 rq->rx_stats.rx_ucast_pkts++;
1809 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1812 struct oce_packet_desc *pd;
1813 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1815 for (i = 0; i < num_frags; i++) {
1816 if (rq->ring->cidx == rq->ring->pidx) {
1817 device_printf(sc->dev,
1818 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1821 pd = &rq->pckts[rq->ring->cidx];
1822 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1823 bus_dmamap_unload(rq->tag, pd->map);
1824 if (pd->mbuf != NULL) {
1829 RING_GET(rq->ring, 1);
1836 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1838 struct oce_nic_rx_cqe_v1 *cqe_v1;
1841 if (sc->be3_native) {
1842 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1843 vtp = cqe_v1->u0.s.vlan_tag_present;
1845 vtp = cqe->u0.s.vlan_tag_present;
1853 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1855 struct oce_nic_rx_cqe_v1 *cqe_v1;
1858 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1859 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1860 port_id = cqe_v1->u0.s.port;
1861 if (sc->port_id != port_id)
1864 ;/* For BE3 legacy and Lancer this is dummy */
1870 #if defined(INET6) || defined(INET)
1872 oce_rx_flush_lro(struct oce_rq *rq)
1874 struct lro_ctrl *lro = &rq->lro;
1875 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1877 if (!IF_LRO_ENABLED(sc))
1880 tcp_lro_flush_all(lro);
1881 rq->lro_pkts_queued = 0;
1888 oce_init_lro(POCE_SOFTC sc)
1890 struct lro_ctrl *lro = NULL;
1893 for (i = 0; i < sc->nrqs; i++) {
1894 lro = &sc->rq[i]->lro;
1895 rc = tcp_lro_init(lro);
1897 device_printf(sc->dev, "LRO init failed\n");
1908 oce_free_lro(POCE_SOFTC sc)
1910 struct lro_ctrl *lro = NULL;
1913 for (i = 0; i < sc->nrqs; i++) {
1914 lro = &sc->rq[i]->lro;
1922 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1924 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1926 struct oce_packet_desc *pd;
1927 bus_dma_segment_t segs[6];
1928 int nsegs, added = 0;
1929 struct oce_nic_rqe *rqe;
1930 pd_rxulp_db_t rxdb_reg;
1932 uint32_t oce_max_rq_posts = 64;
1934 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1935 for (i = 0; i < count; i++) {
1936 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1938 pd = &rq->pckts[rq->ring->pidx];
1939 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1940 if (pd->mbuf == NULL) {
1941 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1944 pd->mbuf->m_nextpkt = NULL;
1946 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1948 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1951 segs, &nsegs, BUS_DMA_NOWAIT);
1954 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1963 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1965 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1966 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1967 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1968 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1969 RING_PUT(rq->ring, 1);
1973 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1975 for (i = added / oce_max_rq_posts; i > 0; i--) {
1976 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1977 rxdb_reg.bits.qid = rq->rq_id;
1979 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1980 val |= oce_max_rq_posts << 16;
1981 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1983 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1985 added -= oce_max_rq_posts;
1988 rxdb_reg.bits.qid = rq->rq_id;
1989 rxdb_reg.bits.num_posted = added;
1991 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1993 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1995 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
2004 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
2007 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
2008 if(!sc->enable_hwlro) {
2009 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
2010 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
2012 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
2013 oce_alloc_rx_bufs(rq, 64);
2021 oce_rq_handler_lro(void *arg)
2023 struct oce_rq *rq = (struct oce_rq *)arg;
2024 struct oce_cq *cq = rq->cq;
2025 POCE_SOFTC sc = rq->parent;
2026 struct nic_hwlro_singleton_cqe *cqe;
2027 struct nic_hwlro_cqe_part2 *cqe2;
2031 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2032 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2033 while (cqe->valid) {
2034 if(cqe->cqe_type == 0) { /* singleton cqe */
2035 /* we should not get singleton cqe after cqe1 on same rq */
2036 if(rq->cqe_firstpart != NULL) {
2037 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2038 goto exit_rq_handler_lro;
2040 if(cqe->error != 0) {
2041 rq->rx_stats.rxcp_err++;
2042 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2044 oce_rx_lro(rq, cqe, NULL);
2045 rq->rx_stats.rx_compl++;
2047 RING_GET(cq->ring, 1);
2049 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2051 }else if(cqe->cqe_type == 0x1) { /* first part */
2052 /* we should not get cqe1 after cqe1 on same rq */
2053 if(rq->cqe_firstpart != NULL) {
2054 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2055 goto exit_rq_handler_lro;
2057 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2058 RING_GET(cq->ring, 1);
2059 }else if(cqe->cqe_type == 0x2) { /* second part */
2060 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2061 if(cqe2->error != 0) {
2062 rq->rx_stats.rxcp_err++;
2063 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2065 /* We should not get cqe2 without cqe1 */
2066 if(rq->cqe_firstpart == NULL) {
2067 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2068 goto exit_rq_handler_lro;
2070 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2072 rq->rx_stats.rx_compl++;
2073 rq->cqe_firstpart->valid = 0;
2075 rq->cqe_firstpart = NULL;
2077 RING_GET(cq->ring, 1);
2079 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2083 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2084 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2086 oce_check_rx_bufs(sc, num_cqes, rq);
2087 exit_rq_handler_lro:
2088 UNLOCK(&rq->rx_lock);
2092 /* Handle the Completion Queue for receive */
2094 oce_rq_handler(void *arg)
2096 struct oce_rq *rq = (struct oce_rq *)arg;
2097 struct oce_cq *cq = rq->cq;
2098 POCE_SOFTC sc = rq->parent;
2099 struct oce_nic_rx_cqe *cqe;
2103 oce_rq_handler_lro(arg);
2107 bus_dmamap_sync(cq->ring->dma.tag,
2108 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2109 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2110 while (cqe->u0.dw[2]) {
2111 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2113 if (cqe->u0.s.error == 0) {
2116 rq->rx_stats.rxcp_err++;
2117 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2118 /* Post L3/L4 errors to stack.*/
2121 rq->rx_stats.rx_compl++;
2124 #if defined(INET6) || defined(INET)
2125 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2126 oce_rx_flush_lro(rq);
2130 RING_GET(cq->ring, 1);
2131 bus_dmamap_sync(cq->ring->dma.tag,
2132 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2134 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2136 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2140 #if defined(INET6) || defined(INET)
2141 if (IF_LRO_ENABLED(sc))
2142 oce_rx_flush_lro(rq);
2145 oce_check_rx_bufs(sc, num_cqes, rq);
2146 UNLOCK(&rq->rx_lock);
2154 /*****************************************************************************
2155 * Helper function prototypes in this file *
2156 *****************************************************************************/
2159 oce_attach_ifp(POCE_SOFTC sc)
2162 sc->ifp = if_alloc(IFT_ETHER);
2166 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2167 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2168 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2170 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2171 sc->ifp->if_ioctl = oce_ioctl;
2172 sc->ifp->if_start = oce_start;
2173 sc->ifp->if_init = oce_init;
2174 sc->ifp->if_mtu = ETHERMTU;
2175 sc->ifp->if_softc = sc;
2176 #if __FreeBSD_version >= 800000
2177 sc->ifp->if_transmit = oce_multiq_start;
2178 sc->ifp->if_qflush = oce_multiq_flush;
2181 if_initname(sc->ifp,
2182 device_get_name(sc->dev), device_get_unit(sc->dev));
2184 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2185 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2186 IFQ_SET_READY(&sc->ifp->if_snd);
2188 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2189 sc->ifp->if_hwassist |= CSUM_TSO;
2190 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2192 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2193 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2194 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2196 #if defined(INET6) || defined(INET)
2197 sc->ifp->if_capabilities |= IFCAP_TSO;
2198 sc->ifp->if_capabilities |= IFCAP_LRO;
2199 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2202 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2203 sc->ifp->if_baudrate = IF_Gbps(10);
2205 #if __FreeBSD_version >= 1000000
2206 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2207 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2208 sc->ifp->if_hw_tsomaxsegsize = 4096;
2211 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2218 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2220 POCE_SOFTC sc = ifp->if_softc;
2222 if (ifp->if_softc != arg)
2224 if ((vtag == 0) || (vtag > 4095))
2227 sc->vlan_tag[vtag] = 1;
2229 if (sc->vlans_added <= (sc->max_vlans + 1))
2235 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2237 POCE_SOFTC sc = ifp->if_softc;
2239 if (ifp->if_softc != arg)
2241 if ((vtag == 0) || (vtag > 4095))
2244 sc->vlan_tag[vtag] = 0;
2251 * A max of 64 vlans can be configured in BE. If the user configures
2252 * more, place the card in vlan promiscuous mode.
2255 oce_vid_config(POCE_SOFTC sc)
2257 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2258 uint16_t ntags = 0, i;
2261 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2262 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2263 for (i = 0; i < MAX_VLANS; i++) {
2264 if (sc->vlan_tag[i]) {
2265 vtags[ntags].vtag = i;
2270 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2271 vtags, ntags, 1, 0);
2273 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2280 oce_mac_addr_set(POCE_SOFTC sc)
2282 uint32_t old_pmac_id = sc->pmac_id;
2286 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2287 sc->macaddr.size_of_struct);
2291 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2292 sc->if_id, &sc->pmac_id);
2294 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2295 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2296 sc->macaddr.size_of_struct);
2299 device_printf(sc->dev, "Failed update macaddress\n");
2305 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2307 POCE_SOFTC sc = ifp->if_softc;
2308 struct ifreq *ifr = (struct ifreq *)data;
2310 char cookie[32] = {0};
2311 void *priv_data = ifr_data_get_ptr(ifr);
2315 OCE_DMA_MEM dma_mem;
2316 struct mbx_common_get_cntl_attr *fw_cmd;
2318 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2321 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2324 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2325 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2328 req_size = le32toh(req.u0.req.request_length);
2329 if (req_size > 65536)
2332 req_size += sizeof(struct mbx_hdr);
2333 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2337 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2342 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2348 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2352 firmware is filling all the attributes for this ioctl except
2353 the driver version..so fill it
2355 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2356 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2357 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2358 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2362 oce_dma_free(sc, &dma_mem);
2368 oce_eqd_set_periodic(POCE_SOFTC sc)
2370 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2371 struct oce_aic_obj *aic;
2373 uint64_t now = 0, delta;
2374 int eqd, i, num = 0;
2375 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2379 #define ticks_to_msecs(t) (1000 * (t) / hz)
2381 for (i = 0 ; i < sc->neqs; i++) {
2383 aic = &sc->aic_obj[i];
2384 /* When setting the static eq delay from the user space */
2393 rxpkts = rq->rx_stats.rx_pkts;
2395 tx_reqs = wq->tx_stats.tx_reqs;
2398 if (!aic->ticks || now < aic->ticks ||
2399 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2400 aic->prev_rxpkts = rxpkts;
2401 aic->prev_txreqs = tx_reqs;
2406 delta = ticks_to_msecs(now - aic->ticks);
2408 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2409 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2410 eqd = (pps / 15000) << 2;
2414 /* Make sure that the eq delay is in the known range */
2415 eqd = min(eqd, aic->max_eqd);
2416 eqd = max(eqd, aic->min_eqd);
2418 aic->prev_rxpkts = rxpkts;
2419 aic->prev_txreqs = tx_reqs;
2423 if (eqd != aic->cur_eqd) {
2424 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2425 set_eqd[num].eq_id = eqo->eq_id;
2431 /* Is there atleast one eq that needs to be modified? */
2432 for(i = 0; i < num; i += 8) {
2434 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2436 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2441 static void oce_detect_hw_error(POCE_SOFTC sc)
2444 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2445 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2452 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2453 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2454 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2455 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2458 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2459 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2460 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2461 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2463 ue_low = (ue_low & ~ue_low_mask);
2464 ue_high = (ue_high & ~ue_high_mask);
2467 /* On certain platforms BE hardware can indicate spurious UEs.
2468 * Allow the h/w to stop working completely in case of a real UE.
2469 * Hence not setting the hw_error for UE detection.
2471 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2472 sc->hw_error = TRUE;
2473 device_printf(sc->dev, "Error detected in the card\n");
2476 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2477 device_printf(sc->dev,
2478 "ERR: sliport status 0x%x\n", sliport_status);
2479 device_printf(sc->dev,
2480 "ERR: sliport error1 0x%x\n", sliport_err1);
2481 device_printf(sc->dev,
2482 "ERR: sliport error2 0x%x\n", sliport_err2);
2486 for (i = 0; ue_low; ue_low >>= 1, i++) {
2488 device_printf(sc->dev, "UE: %s bit set\n",
2489 ue_status_low_desc[i]);
2494 for (i = 0; ue_high; ue_high >>= 1, i++) {
2496 device_printf(sc->dev, "UE: %s bit set\n",
2497 ue_status_hi_desc[i]);
2505 oce_local_timer(void *arg)
2507 POCE_SOFTC sc = arg;
2510 oce_detect_hw_error(sc);
2511 oce_refresh_nic_stats(sc);
2512 oce_refresh_queue_stats(sc);
2513 oce_mac_addr_set(sc);
2516 for (i = 0; i < sc->nwqs; i++)
2517 oce_tx_restart(sc, sc->wq[i]);
2519 /* calculate and set the eq delay for optimal interrupt rate */
2520 if (IS_BE(sc) || IS_SH(sc))
2521 oce_eqd_set_periodic(sc);
2523 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2527 oce_tx_compl_clean(POCE_SOFTC sc)
2530 int i = 0, timeo = 0, num_wqes = 0;
2531 int pending_txqs = sc->nwqs;
2533 /* Stop polling for compls when HW has been silent for 10ms or
2534 * hw_error or no outstanding completions expected
2537 pending_txqs = sc->nwqs;
2539 for_all_wq_queues(sc, wq, i) {
2540 num_wqes = oce_wq_handler(wq);
2545 if(!wq->ring->num_used)
2549 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2555 for_all_wq_queues(sc, wq, i) {
2556 while(wq->ring->num_used) {
2557 LOCK(&wq->tx_compl_lock);
2558 oce_process_tx_completion(wq);
2559 UNLOCK(&wq->tx_compl_lock);
2565 /* NOTE : This should only be called holding
2569 oce_if_deactivate(POCE_SOFTC sc)
2576 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2578 oce_tx_compl_clean(sc);
2580 /* Stop intrs and finish any bottom halves pending */
2581 oce_hw_intr_disable(sc);
2583 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2584 any other lock. So unlock device lock and require after
2585 completing taskqueue_drain.
2587 UNLOCK(&sc->dev_lock);
2588 for (i = 0; i < sc->intr_count; i++) {
2589 if (sc->intrs[i].tq != NULL) {
2590 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2593 LOCK(&sc->dev_lock);
2595 /* Delete RX queue in card with flush param */
2598 /* Invalidate any pending cq and eq entries*/
2599 for_all_evnt_queues(sc, eq, i)
2601 for_all_rq_queues(sc, rq, i)
2602 oce_drain_rq_cq(rq);
2603 for_all_wq_queues(sc, wq, i)
2604 oce_drain_wq_cq(wq);
2606 /* But still we need to get MCC aync events.
2607 So enable intrs and also arm first EQ
2609 oce_hw_intr_enable(sc);
2610 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2617 oce_if_activate(POCE_SOFTC sc)
2624 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2626 oce_hw_intr_disable(sc);
2630 for_all_rq_queues(sc, rq, i) {
2631 rc = oce_start_rq(rq);
2633 device_printf(sc->dev, "Unable to start RX\n");
2636 for_all_wq_queues(sc, wq, i) {
2637 rc = oce_start_wq(wq);
2639 device_printf(sc->dev, "Unable to start TX\n");
2643 for_all_evnt_queues(sc, eq, i)
2644 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2646 oce_hw_intr_enable(sc);
2651 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2653 /* Update Link status */
2654 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2655 ASYNC_EVENT_LINK_UP) {
2656 sc->link_status = ASYNC_EVENT_LINK_UP;
2657 if_link_state_change(sc->ifp, LINK_STATE_UP);
2659 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2660 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2665 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2666 struct oce_async_evt_grp5_os2bmc *evt)
2668 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2669 if (evt->u.s.mgmt_enable)
2670 sc->flags |= OCE_FLAGS_OS2BMC;
2674 sc->bmc_filt_mask = evt->u.s.arp_filter;
2675 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2676 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2677 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2678 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2679 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2680 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2681 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2682 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2686 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2688 struct oce_async_event_grp5_pvid_state *gcqe;
2689 struct oce_async_evt_grp5_os2bmc *bmccqe;
2691 switch (cqe->u0.s.async_type) {
2692 case ASYNC_EVENT_PVID_STATE:
2694 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2696 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2700 case ASYNC_EVENT_OS2BMC:
2701 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2702 oce_async_grp5_osbmc_process(sc, bmccqe);
2709 /* Handle the Completion Queue for the Mailbox/Async notifications */
2711 oce_mq_handler(void *arg)
2713 struct oce_mq *mq = (struct oce_mq *)arg;
2714 POCE_SOFTC sc = mq->parent;
2715 struct oce_cq *cq = mq->cq;
2716 int num_cqes = 0, evt_type = 0, optype = 0;
2717 struct oce_mq_cqe *cqe;
2718 struct oce_async_cqe_link_state *acqe;
2719 struct oce_async_event_qnq *dbgcqe;
2722 bus_dmamap_sync(cq->ring->dma.tag,
2723 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2724 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2726 while (cqe->u0.dw[3]) {
2727 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2728 if (cqe->u0.s.async_event) {
2729 evt_type = cqe->u0.s.event_type;
2730 optype = cqe->u0.s.async_type;
2731 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2732 /* Link status evt */
2733 acqe = (struct oce_async_cqe_link_state *)cqe;
2734 process_link_state(sc, acqe);
2735 } else if (evt_type == ASYNC_EVENT_GRP5) {
2736 oce_process_grp5_events(sc, cqe);
2737 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2738 optype == ASYNC_EVENT_DEBUG_QNQ) {
2739 dbgcqe = (struct oce_async_event_qnq *)cqe;
2741 sc->qnqid = dbgcqe->vlan_tag;
2742 sc->qnq_debug_event = TRUE;
2746 RING_GET(cq->ring, 1);
2747 bus_dmamap_sync(cq->ring->dma.tag,
2748 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2749 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2754 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2761 setup_max_queues_want(POCE_SOFTC sc)
2763 /* Check if it is FLEX machine. Is so dont use RSS */
2764 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2765 (sc->function_mode & FNM_UMC_MODE) ||
2766 (sc->function_mode & FNM_VNIC_MODE) ||
2767 (!is_rss_enabled(sc)) ||
2772 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2773 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2776 if (IS_BE2(sc) && is_rss_enabled(sc))
2777 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2782 update_queues_got(POCE_SOFTC sc)
2784 if (is_rss_enabled(sc)) {
2785 sc->nrqs = sc->intr_count + 1;
2786 sc->nwqs = sc->intr_count;
2797 oce_check_ipv6_ext_hdr(struct mbuf *m)
2799 struct ether_header *eh = mtod(m, struct ether_header *);
2800 caddr_t m_datatemp = m->m_data;
2802 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2803 m->m_data += sizeof(struct ether_header);
2804 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2806 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2807 (ip6->ip6_nxt != IPPROTO_UDP)){
2808 struct ip6_ext *ip6e = NULL;
2809 m->m_data += sizeof(struct ip6_hdr);
2811 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2812 if(ip6e->ip6e_len == 0xff) {
2813 m->m_data = m_datatemp;
2817 m->m_data = m_datatemp;
2823 is_be3_a1(POCE_SOFTC sc)
2825 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2831 static struct mbuf *
2832 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2834 uint16_t vlan_tag = 0;
2839 /* Embed vlan tag in the packet if it is not part of it */
2840 if(m->m_flags & M_VLANTAG) {
2841 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2842 m->m_flags &= ~M_VLANTAG;
2845 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2848 vlan_tag = sc->pvid;
2854 m = ether_vlanencap(m, vlan_tag);
2858 m = ether_vlanencap(m, sc->qnqid);
2867 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2869 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2870 oce_check_ipv6_ext_hdr(m)) {
2877 oce_get_config(POCE_SOFTC sc)
2880 uint32_t max_rss = 0;
2882 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2883 max_rss = OCE_LEGACY_MODE_RSS;
2885 max_rss = OCE_MAX_RSS;
2888 rc = oce_get_profile_config(sc, max_rss);
2890 sc->nwqs = OCE_MAX_WQ;
2891 sc->nrssqs = max_rss;
2892 sc->nrqs = sc->nrssqs + 1;
2895 else { /* For BE3 don't rely on fw for determining the resources */
2896 sc->nrssqs = max_rss;
2897 sc->nrqs = sc->nrssqs + 1;
2898 sc->nwqs = OCE_MAX_WQ;
2899 sc->max_vlans = MAX_VLANFILTER_SIZE;
2904 oce_rdma_close(void)
2906 if (oce_rdma_if != NULL) {
2912 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2914 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2918 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2921 struct oce_dev_info di;
2924 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2928 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2929 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2933 rdma_info->close = oce_rdma_close;
2934 rdma_info->mbox_post = oce_mbox_post;
2935 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2936 rdma_info->get_mac_addr = oce_get_mac_addr;
2938 oce_rdma_if = rdma_if;
2941 while (sc != NULL) {
2942 if (oce_rdma_if->announce != NULL) {
2943 memset(&di, 0, sizeof(di));
2947 di.db_bhandle = sc->db_bhandle;
2948 di.db_btag = sc->db_btag;
2949 di.db_page_size = 4096;
2950 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2951 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2952 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2953 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2955 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2957 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2958 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2959 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2960 di.msix.start_vector = sc->intr_count;
2961 for (i=0; i<di.msix.num_vectors; i++) {
2962 di.msix.vector_list[i] = sc->intrs[i].vector;
2966 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2967 di.vendor_id = pci_get_vendor(sc->dev);
2968 di.dev_id = pci_get_device(sc->dev);
2970 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2971 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2974 rdma_if->announce(&di);
2983 oce_read_env_variables( POCE_SOFTC sc )
2988 /* read if user wants to enable hwlro or swlro */
2989 //value = getenv("oce_enable_hwlro");
2990 if(value && IS_SH(sc)) {
2991 sc->enable_hwlro = strtol(value, NULL, 10);
2992 if(sc->enable_hwlro) {
2993 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
2995 device_printf(sc->dev, "no hardware lro support\n");
2996 device_printf(sc->dev, "software lro enabled\n");
2997 sc->enable_hwlro = 0;
2999 device_printf(sc->dev, "hardware lro enabled\n");
3000 oce_max_rsp_handled = 32;
3003 device_printf(sc->dev, "software lro enabled\n");
3006 sc->enable_hwlro = 0;
3009 /* read mbuf size */
3010 //value = getenv("oce_rq_buf_size");
3011 if(value && IS_SH(sc)) {
3012 oce_rq_buf_size = strtol(value, NULL, 10);
3013 switch(oce_rq_buf_size) {
3021 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
3022 oce_rq_buf_size = 2048;