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;
486 rc = ifmedia_ioctl(ifp, ifr, &sc->media, command);
490 if (ifr->ifr_mtu > OCE_MAX_MTU)
493 ifp->if_mtu = ifr->ifr_mtu;
497 if (ifp->if_flags & IFF_UP) {
498 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
499 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
502 device_printf(sc->dev, "Interface Up\n");
506 sc->ifp->if_drv_flags &=
507 ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
508 oce_if_deactivate(sc);
510 UNLOCK(&sc->dev_lock);
512 device_printf(sc->dev, "Interface Down\n");
515 if ((ifp->if_flags & IFF_PROMISC) && !sc->promisc) {
516 if (!oce_rxf_set_promiscuous(sc, (1 | (1 << 1))))
518 } else if (!(ifp->if_flags & IFF_PROMISC) && sc->promisc) {
519 if (!oce_rxf_set_promiscuous(sc, 0))
527 rc = oce_hw_update_multicast(sc);
529 device_printf(sc->dev,
530 "Update multicast address failed\n");
534 u = ifr->ifr_reqcap ^ ifp->if_capenable;
536 if (u & IFCAP_TXCSUM) {
537 ifp->if_capenable ^= IFCAP_TXCSUM;
538 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
540 if (IFCAP_TSO & ifp->if_capenable &&
541 !(IFCAP_TXCSUM & ifp->if_capenable)) {
542 ifp->if_capenable &= ~IFCAP_TSO;
543 ifp->if_hwassist &= ~CSUM_TSO;
545 "TSO disabled due to -txcsum.\n");
549 if (u & IFCAP_RXCSUM)
550 ifp->if_capenable ^= IFCAP_RXCSUM;
552 if (u & IFCAP_TSO4) {
553 ifp->if_capenable ^= IFCAP_TSO4;
555 if (IFCAP_TSO & ifp->if_capenable) {
556 if (IFCAP_TXCSUM & ifp->if_capenable)
557 ifp->if_hwassist |= CSUM_TSO;
559 ifp->if_capenable &= ~IFCAP_TSO;
560 ifp->if_hwassist &= ~CSUM_TSO;
562 "Enable txcsum first.\n");
566 ifp->if_hwassist &= ~CSUM_TSO;
569 if (u & IFCAP_VLAN_HWTAGGING)
570 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
572 if (u & IFCAP_VLAN_HWFILTER) {
573 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
576 #if defined(INET6) || defined(INET)
578 ifp->if_capenable ^= IFCAP_LRO;
579 if(sc->enable_hwlro) {
580 if(ifp->if_capenable & IFCAP_LRO) {
581 rc = oce_mbox_nic_set_iface_lro_config(sc, 1);
583 rc = oce_mbox_nic_set_iface_lro_config(sc, 0);
592 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
596 if (i2c.dev_addr == PAGE_NUM_A0) {
598 } else if (i2c.dev_addr == PAGE_NUM_A2) {
599 offset = TRANSCEIVER_A0_SIZE + i2c.offset;
605 if (i2c.len > sizeof(i2c.data) ||
606 i2c.len + offset > sizeof(sfp_vpd_dump_buffer)) {
611 rc = oce_mbox_read_transrecv_data(sc, i2c.dev_addr);
617 memcpy(&i2c.data[0], &sfp_vpd_dump_buffer[offset], i2c.len);
619 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
623 rc = priv_check(curthread, PRIV_DRIVER);
626 rc = oce_handle_passthrough(ifp, data);
629 rc = ether_ioctl(ifp, command, data);
644 if (sc->ifp->if_flags & IFF_UP) {
645 oce_if_deactivate(sc);
649 UNLOCK(&sc->dev_lock);
655 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
657 POCE_SOFTC sc = ifp->if_softc;
658 struct oce_wq *wq = NULL;
662 if (!sc->link_status)
665 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
666 queue_index = m->m_pkthdr.flowid % sc->nwqs;
668 wq = sc->wq[queue_index];
671 status = oce_multiq_transmit(ifp, m, wq);
672 UNLOCK(&wq->tx_lock);
680 oce_multiq_flush(struct ifnet *ifp)
682 POCE_SOFTC sc = ifp->if_softc;
686 for (i = 0; i < sc->nwqs; i++) {
687 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
695 /*****************************************************************************
696 * Driver interrupt routines functions *
697 *****************************************************************************/
700 oce_intr(void *arg, int pending)
703 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
704 POCE_SOFTC sc = ii->sc;
705 struct oce_eq *eq = ii->eq;
707 struct oce_cq *cq = NULL;
711 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
712 BUS_DMASYNC_POSTWRITE);
714 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
718 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
719 BUS_DMASYNC_POSTWRITE);
720 RING_GET(eq->ring, 1);
726 goto eq_arm; /* Spurious */
728 /* Clear EQ entries, but dont arm */
729 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
731 /* Process TX, RX and MCC. But dont arm CQ*/
732 for (i = 0; i < eq->cq_valid; i++) {
734 (*cq->cq_handler)(cq->cb_arg);
737 /* Arm all cqs connected to this EQ */
738 for (i = 0; i < eq->cq_valid; i++) {
740 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
744 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
751 oce_setup_intr(POCE_SOFTC sc)
753 int rc = 0, use_intx = 0;
754 int vector = 0, req_vectors = 0;
755 int tot_req_vectors, tot_vectors;
757 if (is_rss_enabled(sc))
758 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
762 tot_req_vectors = req_vectors;
763 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
764 if (req_vectors > 1) {
765 tot_req_vectors += OCE_RDMA_VECTORS;
766 sc->roce_intr_count = OCE_RDMA_VECTORS;
770 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
771 sc->intr_count = req_vectors;
772 tot_vectors = tot_req_vectors;
773 rc = pci_alloc_msix(sc->dev, &tot_vectors);
776 pci_release_msi(sc->dev);
778 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
779 if (tot_vectors < tot_req_vectors) {
780 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
781 sc->roce_intr_count = (tot_vectors / 2);
783 sc->intr_count = tot_vectors - sc->roce_intr_count;
786 sc->intr_count = tot_vectors;
788 sc->flags |= OCE_FLAGS_USING_MSIX;
796 /* Scale number of queues based on intr we got */
797 update_queues_got(sc);
800 device_printf(sc->dev, "Using legacy interrupt\n");
801 rc = oce_alloc_intr(sc, vector, oce_intr);
805 for (; vector < sc->intr_count; vector++) {
806 rc = oce_alloc_intr(sc, vector, oce_intr);
820 oce_fast_isr(void *arg)
822 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
823 POCE_SOFTC sc = ii->sc;
828 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
830 taskqueue_enqueue(ii->tq, &ii->task);
834 return FILTER_HANDLED;
839 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
844 if (vector >= OCE_MAX_EQ)
847 ii = &sc->intrs[vector];
849 /* Set the resource id for the interrupt.
850 * MSIx is vector + 1 for the resource id,
851 * INTx is 0 for the resource id.
853 if (sc->flags & OCE_FLAGS_USING_MSIX)
857 ii->intr_res = bus_alloc_resource_any(sc->dev,
859 &rr, RF_ACTIVE|RF_SHAREABLE);
861 if (ii->intr_res == NULL) {
862 device_printf(sc->dev,
863 "Could not allocate interrupt\n");
868 TASK_INIT(&ii->task, 0, isr, ii);
870 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
871 ii->tq = taskqueue_create_fast(ii->task_name,
873 taskqueue_thread_enqueue,
875 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
876 device_get_nameunit(sc->dev));
879 rc = bus_setup_intr(sc->dev,
882 oce_fast_isr, NULL, ii, &ii->tag);
889 oce_intr_free(POCE_SOFTC sc)
893 for (i = 0; i < sc->intr_count; i++) {
895 if (sc->intrs[i].tag != NULL)
896 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
898 if (sc->intrs[i].tq != NULL)
899 taskqueue_free(sc->intrs[i].tq);
901 if (sc->intrs[i].intr_res != NULL)
902 bus_release_resource(sc->dev, SYS_RES_IRQ,
904 sc->intrs[i].intr_res);
905 sc->intrs[i].tag = NULL;
906 sc->intrs[i].intr_res = NULL;
909 if (sc->flags & OCE_FLAGS_USING_MSIX)
910 pci_release_msi(sc->dev);
916 /******************************************************************************
917 * Media callbacks functions *
918 ******************************************************************************/
921 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
923 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
926 req->ifm_status = IFM_AVALID;
927 req->ifm_active = IFM_ETHER;
929 if (sc->link_status == 1)
930 req->ifm_status |= IFM_ACTIVE;
934 switch (sc->link_speed) {
935 case 1: /* 10 Mbps */
936 req->ifm_active |= IFM_10_T | IFM_FDX;
939 case 2: /* 100 Mbps */
940 req->ifm_active |= IFM_100_TX | IFM_FDX;
944 req->ifm_active |= IFM_1000_T | IFM_FDX;
947 case 4: /* 10 Gbps */
948 req->ifm_active |= IFM_10G_SR | IFM_FDX;
951 case 5: /* 20 Gbps */
952 req->ifm_active |= IFM_10G_SR | IFM_FDX;
955 case 6: /* 25 Gbps */
956 req->ifm_active |= IFM_10G_SR | IFM_FDX;
959 case 7: /* 40 Gbps */
960 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
973 oce_media_change(struct ifnet *ifp)
979 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
980 struct mbuf *m, boolean_t *os2bmc,
983 struct ether_header *eh = NULL;
985 eh = mtod(m, struct ether_header *);
987 if (!is_os2bmc_enabled(sc) || *os2bmc) {
991 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
994 if (is_mc_allowed_on_bmc(sc, eh) ||
995 is_bc_allowed_on_bmc(sc, eh) ||
996 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
1001 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
1002 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1003 uint8_t nexthdr = ip6->ip6_nxt;
1004 if (nexthdr == IPPROTO_ICMPV6) {
1005 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
1006 switch (icmp6->icmp6_type) {
1007 case ND_ROUTER_ADVERT:
1008 *os2bmc = is_ipv6_ra_filt_enabled(sc);
1010 case ND_NEIGHBOR_ADVERT:
1011 *os2bmc = is_ipv6_na_filt_enabled(sc);
1019 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
1020 struct ip *ip = mtod(m, struct ip *);
1021 int iphlen = ip->ip_hl << 2;
1022 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
1023 switch (uh->uh_dport) {
1024 case DHCP_CLIENT_PORT:
1025 *os2bmc = is_dhcp_client_filt_enabled(sc);
1027 case DHCP_SERVER_PORT:
1028 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
1030 case NET_BIOS_PORT1:
1031 case NET_BIOS_PORT2:
1032 *os2bmc = is_nbios_filt_enabled(sc);
1034 case DHCPV6_RAS_PORT:
1035 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1043 *m_new = m_dup(m, M_NOWAIT);
1048 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1054 /*****************************************************************************
1055 * Transmit routines functions *
1056 *****************************************************************************/
1059 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1061 int rc = 0, i, retry_cnt = 0;
1062 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1063 struct mbuf *m, *m_temp, *m_new = NULL;
1064 struct oce_wq *wq = sc->wq[wq_index];
1065 struct oce_packet_desc *pd;
1066 struct oce_nic_hdr_wqe *nichdr;
1067 struct oce_nic_frag_wqe *nicfrag;
1068 struct ether_header *eh = NULL;
1071 boolean_t complete = TRUE;
1072 boolean_t os2bmc = FALSE;
1078 if (!(m->m_flags & M_PKTHDR)) {
1083 /* Don't allow non-TSO packets longer than MTU */
1084 if (!is_tso_pkt(m)) {
1085 eh = mtod(m, struct ether_header *);
1086 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1090 if(oce_tx_asic_stall_verify(sc, m)) {
1091 m = oce_insert_vlan_tag(sc, m, &complete);
1093 device_printf(sc->dev, "Insertion unsuccessful\n");
1099 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1100 * may cause a transmit stall on that port. So the work-around is to
1101 * pad short packets (<= 32 bytes) to a 36-byte length.
1103 if(IS_SH(sc) || IS_XE201(sc) ) {
1104 if(m->m_pkthdr.len <= 32) {
1106 bzero((void *)buf, 36);
1107 m_append(m, (36 - m->m_pkthdr.len), buf);
1112 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1113 /* consolidate packet buffers for TSO/LSO segment offload */
1114 #if defined(INET6) || defined(INET)
1115 m = oce_tso_setup(sc, mpp);
1126 pd = &wq->pckts[wq->pkt_desc_head];
1129 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1131 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1133 num_wqes = pd->nsegs + 1;
1134 if (IS_BE(sc) || IS_SH(sc)) {
1135 /*Dummy required only for BE3.*/
1139 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1140 bus_dmamap_unload(wq->tag, pd->map);
1143 atomic_store_rel_int(&wq->pkt_desc_head,
1144 (wq->pkt_desc_head + 1) % \
1145 OCE_WQ_PACKET_ARRAY_SIZE);
1146 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1150 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1151 nichdr->u0.dw[0] = 0;
1152 nichdr->u0.dw[1] = 0;
1153 nichdr->u0.dw[2] = 0;
1154 nichdr->u0.dw[3] = 0;
1156 nichdr->u0.s.complete = complete;
1157 nichdr->u0.s.mgmt = os2bmc;
1158 nichdr->u0.s.event = 1;
1159 nichdr->u0.s.crc = 1;
1160 nichdr->u0.s.forward = 0;
1161 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1162 nichdr->u0.s.udpcs =
1163 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1164 nichdr->u0.s.tcpcs =
1165 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1166 nichdr->u0.s.num_wqe = num_wqes;
1167 nichdr->u0.s.total_length = m->m_pkthdr.len;
1169 if (m->m_flags & M_VLANTAG) {
1170 nichdr->u0.s.vlan = 1; /*Vlan present*/
1171 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1174 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1175 if (m->m_pkthdr.tso_segsz) {
1176 nichdr->u0.s.lso = 1;
1177 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1179 if (!IS_BE(sc) || !IS_SH(sc))
1180 nichdr->u0.s.ipcs = 1;
1183 RING_PUT(wq->ring, 1);
1184 atomic_add_int(&wq->ring->num_used, 1);
1186 for (i = 0; i < pd->nsegs; i++) {
1188 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1189 struct oce_nic_frag_wqe);
1190 nicfrag->u0.s.rsvd0 = 0;
1191 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1192 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1193 nicfrag->u0.s.frag_len = segs[i].ds_len;
1194 pd->wqe_idx = wq->ring->pidx;
1195 RING_PUT(wq->ring, 1);
1196 atomic_add_int(&wq->ring->num_used, 1);
1198 if (num_wqes > (pd->nsegs + 1)) {
1200 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1201 struct oce_nic_frag_wqe);
1202 nicfrag->u0.dw[0] = 0;
1203 nicfrag->u0.dw[1] = 0;
1204 nicfrag->u0.dw[2] = 0;
1205 nicfrag->u0.dw[3] = 0;
1206 pd->wqe_idx = wq->ring->pidx;
1207 RING_PUT(wq->ring, 1);
1208 atomic_add_int(&wq->ring->num_used, 1);
1212 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1213 wq->tx_stats.tx_reqs++;
1214 wq->tx_stats.tx_wrbs += num_wqes;
1215 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1216 wq->tx_stats.tx_pkts++;
1218 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1219 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1220 reg_value = (num_wqes << 16) | wq->wq_id;
1222 /* if os2bmc is not enabled or if the pkt is already tagged as
1225 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1227 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1229 } else if (rc == EFBIG) {
1230 if (retry_cnt == 0) {
1231 m_temp = m_defrag(m, M_NOWAIT);
1236 retry_cnt = retry_cnt + 1;
1240 } else if (rc == ENOMEM)
1260 oce_process_tx_completion(struct oce_wq *wq)
1262 struct oce_packet_desc *pd;
1263 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1266 pd = &wq->pckts[wq->pkt_desc_tail];
1267 atomic_store_rel_int(&wq->pkt_desc_tail,
1268 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1269 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1270 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1271 bus_dmamap_unload(wq->tag, pd->map);
1278 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1279 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1280 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1281 oce_tx_restart(sc, wq);
1288 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1291 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1294 if (!drbr_empty(sc->ifp, wq->br))
1295 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1300 #if defined(INET6) || defined(INET)
1301 static struct mbuf *
1302 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1309 struct ip6_hdr *ip6;
1311 struct ether_vlan_header *eh;
1314 int total_len = 0, ehdrlen = 0;
1318 if (M_WRITABLE(m) == 0) {
1319 m = m_dup(*mpp, M_NOWAIT);
1326 eh = mtod(m, struct ether_vlan_header *);
1327 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1328 etype = ntohs(eh->evl_proto);
1329 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1331 etype = ntohs(eh->evl_encap_proto);
1332 ehdrlen = ETHER_HDR_LEN;
1338 ip = (struct ip *)(m->m_data + ehdrlen);
1339 if (ip->ip_p != IPPROTO_TCP)
1341 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1343 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1347 case ETHERTYPE_IPV6:
1348 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1349 if (ip6->ip6_nxt != IPPROTO_TCP)
1351 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1353 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1360 m = m_pullup(m, total_len);
1367 #endif /* INET6 || INET */
1370 oce_tx_task(void *arg, int npending)
1372 struct oce_wq *wq = arg;
1373 POCE_SOFTC sc = wq->parent;
1374 struct ifnet *ifp = sc->ifp;
1378 rc = oce_multiq_transmit(ifp, NULL, wq);
1380 device_printf(sc->dev,
1381 "TX[%d] restart failed\n", wq->queue_index);
1383 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 (rq->queue_index)
1671 m->m_pkthdr.flowid = (rq->queue_index - 1);
1673 m->m_pkthdr.flowid = rq->queue_index;
1674 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1676 /* This deternies if vlan tag is Valid */
1678 if (sc->function_mode & FNM_FLEX10_MODE) {
1679 /* FLEX10. If QnQ is not set, neglect VLAN */
1681 m->m_pkthdr.ether_vtag = cq_info.vtag;
1682 m->m_flags |= M_VLANTAG;
1684 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1685 /* In UMC mode generally pvid will be striped by
1686 hw. But in some cases we have seen it comes
1687 with pvid. So if pvid == vlan, neglect vlan.
1689 m->m_pkthdr.ether_vtag = cq_info.vtag;
1690 m->m_flags |= M_VLANTAG;
1693 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1695 (*sc->ifp->if_input) (sc->ifp, m);
1697 /* Update rx stats per queue */
1698 rq->rx_stats.rx_pkts++;
1699 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1700 rq->rx_stats.rx_frags += cq_info.num_frags;
1701 rq->rx_stats.rx_ucast_pkts++;
1707 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1709 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1711 struct mbuf *m = NULL;
1712 struct oce_common_cqe_info cq_info;
1715 /* Is it a flush compl that has no data */
1716 if(!cqe->u0.s.num_fragments)
1719 len = cqe->u0.s.pkt_size;
1721 /*partial DMA workaround for Lancer*/
1722 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1726 if (!oce_cqe_portid_valid(sc, cqe)) {
1727 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1731 /* Get vlan_tag value */
1732 if(IS_BE(sc) || IS_SH(sc))
1733 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1735 vtag = cqe->u0.s.vlan_tag;
1737 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1738 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1739 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1740 cq_info.num_frags = cqe->u0.s.num_fragments;
1741 cq_info.pkt_size = cqe->u0.s.pkt_size;
1743 oce_rx_mbuf_chain(rq, &cq_info, &m);
1746 m->m_pkthdr.rcvif = sc->ifp;
1747 if (rq->queue_index)
1748 m->m_pkthdr.flowid = (rq->queue_index - 1);
1750 m->m_pkthdr.flowid = rq->queue_index;
1751 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1753 /* This deternies if vlan tag is Valid */
1754 if (oce_cqe_vtp_valid(sc, cqe)) {
1755 if (sc->function_mode & FNM_FLEX10_MODE) {
1756 /* FLEX10. If QnQ is not set, neglect VLAN */
1757 if (cqe->u0.s.qnq) {
1758 m->m_pkthdr.ether_vtag = vtag;
1759 m->m_flags |= M_VLANTAG;
1761 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1762 /* In UMC mode generally pvid will be striped by
1763 hw. But in some cases we have seen it comes
1764 with pvid. So if pvid == vlan, neglect vlan.
1766 m->m_pkthdr.ether_vtag = vtag;
1767 m->m_flags |= M_VLANTAG;
1771 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1772 #if defined(INET6) || defined(INET)
1773 /* Try to queue to LRO */
1774 if (IF_LRO_ENABLED(sc) &&
1775 (cqe->u0.s.ip_cksum_pass) &&
1776 (cqe->u0.s.l4_cksum_pass) &&
1777 (!cqe->u0.s.ip_ver) &&
1778 (rq->lro.lro_cnt != 0)) {
1780 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1781 rq->lro_pkts_queued ++;
1784 /* If LRO posting fails then try to post to STACK */
1788 (*sc->ifp->if_input) (sc->ifp, m);
1789 #if defined(INET6) || defined(INET)
1792 /* Update rx stats per queue */
1793 rq->rx_stats.rx_pkts++;
1794 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1795 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1796 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1797 rq->rx_stats.rx_mcast_pkts++;
1798 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1799 rq->rx_stats.rx_ucast_pkts++;
1807 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1810 struct oce_packet_desc *pd;
1811 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1813 for (i = 0; i < num_frags; i++) {
1814 if (rq->ring->cidx == rq->ring->pidx) {
1815 device_printf(sc->dev,
1816 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1819 pd = &rq->pckts[rq->ring->cidx];
1820 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1821 bus_dmamap_unload(rq->tag, pd->map);
1822 if (pd->mbuf != NULL) {
1827 RING_GET(rq->ring, 1);
1834 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1836 struct oce_nic_rx_cqe_v1 *cqe_v1;
1839 if (sc->be3_native) {
1840 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1841 vtp = cqe_v1->u0.s.vlan_tag_present;
1843 vtp = cqe->u0.s.vlan_tag_present;
1851 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1853 struct oce_nic_rx_cqe_v1 *cqe_v1;
1856 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1857 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1858 port_id = cqe_v1->u0.s.port;
1859 if (sc->port_id != port_id)
1862 ;/* For BE3 legacy and Lancer this is dummy */
1868 #if defined(INET6) || defined(INET)
1870 oce_rx_flush_lro(struct oce_rq *rq)
1872 struct lro_ctrl *lro = &rq->lro;
1873 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1875 if (!IF_LRO_ENABLED(sc))
1878 tcp_lro_flush_all(lro);
1879 rq->lro_pkts_queued = 0;
1886 oce_init_lro(POCE_SOFTC sc)
1888 struct lro_ctrl *lro = NULL;
1891 for (i = 0; i < sc->nrqs; i++) {
1892 lro = &sc->rq[i]->lro;
1893 rc = tcp_lro_init(lro);
1895 device_printf(sc->dev, "LRO init failed\n");
1906 oce_free_lro(POCE_SOFTC sc)
1908 struct lro_ctrl *lro = NULL;
1911 for (i = 0; i < sc->nrqs; i++) {
1912 lro = &sc->rq[i]->lro;
1920 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1922 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1924 struct oce_packet_desc *pd;
1925 bus_dma_segment_t segs[6];
1926 int nsegs, added = 0;
1927 struct oce_nic_rqe *rqe;
1928 pd_rxulp_db_t rxdb_reg;
1930 uint32_t oce_max_rq_posts = 64;
1932 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1933 for (i = 0; i < count; i++) {
1934 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1936 pd = &rq->pckts[rq->ring->pidx];
1937 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1938 if (pd->mbuf == NULL) {
1939 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1942 pd->mbuf->m_nextpkt = NULL;
1944 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1946 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1949 segs, &nsegs, BUS_DMA_NOWAIT);
1952 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1961 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1963 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1964 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1965 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1966 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1967 RING_PUT(rq->ring, 1);
1971 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1973 for (i = added / oce_max_rq_posts; i > 0; i--) {
1974 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1975 rxdb_reg.bits.qid = rq->rq_id;
1977 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1978 val |= oce_max_rq_posts << 16;
1979 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1981 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1983 added -= oce_max_rq_posts;
1986 rxdb_reg.bits.qid = rq->rq_id;
1987 rxdb_reg.bits.num_posted = added;
1989 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1991 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1993 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
2002 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
2005 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
2006 if(!sc->enable_hwlro) {
2007 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
2008 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
2010 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
2011 oce_alloc_rx_bufs(rq, 64);
2019 oce_rq_handler_lro(void *arg)
2021 struct oce_rq *rq = (struct oce_rq *)arg;
2022 struct oce_cq *cq = rq->cq;
2023 POCE_SOFTC sc = rq->parent;
2024 struct nic_hwlro_singleton_cqe *cqe;
2025 struct nic_hwlro_cqe_part2 *cqe2;
2029 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2030 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2031 while (cqe->valid) {
2032 if(cqe->cqe_type == 0) { /* singleton cqe */
2033 /* we should not get singleton cqe after cqe1 on same rq */
2034 if(rq->cqe_firstpart != NULL) {
2035 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2036 goto exit_rq_handler_lro;
2038 if(cqe->error != 0) {
2039 rq->rx_stats.rxcp_err++;
2040 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2042 oce_rx_lro(rq, cqe, NULL);
2043 rq->rx_stats.rx_compl++;
2045 RING_GET(cq->ring, 1);
2047 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2049 }else if(cqe->cqe_type == 0x1) { /* first part */
2050 /* we should not get cqe1 after cqe1 on same rq */
2051 if(rq->cqe_firstpart != NULL) {
2052 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2053 goto exit_rq_handler_lro;
2055 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2056 RING_GET(cq->ring, 1);
2057 }else if(cqe->cqe_type == 0x2) { /* second part */
2058 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2059 if(cqe2->error != 0) {
2060 rq->rx_stats.rxcp_err++;
2061 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2063 /* We should not get cqe2 without cqe1 */
2064 if(rq->cqe_firstpart == NULL) {
2065 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2066 goto exit_rq_handler_lro;
2068 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2070 rq->rx_stats.rx_compl++;
2071 rq->cqe_firstpart->valid = 0;
2073 rq->cqe_firstpart = NULL;
2075 RING_GET(cq->ring, 1);
2077 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2081 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2082 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2084 oce_check_rx_bufs(sc, num_cqes, rq);
2085 exit_rq_handler_lro:
2086 UNLOCK(&rq->rx_lock);
2090 /* Handle the Completion Queue for receive */
2092 oce_rq_handler(void *arg)
2094 struct oce_rq *rq = (struct oce_rq *)arg;
2095 struct oce_cq *cq = rq->cq;
2096 POCE_SOFTC sc = rq->parent;
2097 struct oce_nic_rx_cqe *cqe;
2101 oce_rq_handler_lro(arg);
2105 bus_dmamap_sync(cq->ring->dma.tag,
2106 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2107 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2108 while (cqe->u0.dw[2]) {
2109 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2111 if (cqe->u0.s.error == 0) {
2114 rq->rx_stats.rxcp_err++;
2115 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2116 /* Post L3/L4 errors to stack.*/
2119 rq->rx_stats.rx_compl++;
2122 #if defined(INET6) || defined(INET)
2123 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2124 oce_rx_flush_lro(rq);
2128 RING_GET(cq->ring, 1);
2129 bus_dmamap_sync(cq->ring->dma.tag,
2130 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2132 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2134 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2138 #if defined(INET6) || defined(INET)
2139 if (IF_LRO_ENABLED(sc))
2140 oce_rx_flush_lro(rq);
2143 oce_check_rx_bufs(sc, num_cqes, rq);
2144 UNLOCK(&rq->rx_lock);
2152 /*****************************************************************************
2153 * Helper function prototypes in this file *
2154 *****************************************************************************/
2157 oce_attach_ifp(POCE_SOFTC sc)
2160 sc->ifp = if_alloc(IFT_ETHER);
2164 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2165 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2166 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2168 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2169 sc->ifp->if_ioctl = oce_ioctl;
2170 sc->ifp->if_start = oce_start;
2171 sc->ifp->if_init = oce_init;
2172 sc->ifp->if_mtu = ETHERMTU;
2173 sc->ifp->if_softc = sc;
2174 sc->ifp->if_transmit = oce_multiq_start;
2175 sc->ifp->if_qflush = oce_multiq_flush;
2177 if_initname(sc->ifp,
2178 device_get_name(sc->dev), device_get_unit(sc->dev));
2180 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2181 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2182 IFQ_SET_READY(&sc->ifp->if_snd);
2184 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2185 sc->ifp->if_hwassist |= CSUM_TSO;
2186 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2188 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2189 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2190 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2192 #if defined(INET6) || defined(INET)
2193 sc->ifp->if_capabilities |= IFCAP_TSO;
2194 sc->ifp->if_capabilities |= IFCAP_LRO;
2195 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2198 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2199 sc->ifp->if_baudrate = IF_Gbps(10);
2201 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2202 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2203 sc->ifp->if_hw_tsomaxsegsize = 4096;
2205 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2212 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2214 POCE_SOFTC sc = ifp->if_softc;
2216 if (ifp->if_softc != arg)
2218 if ((vtag == 0) || (vtag > 4095))
2221 sc->vlan_tag[vtag] = 1;
2223 if (sc->vlans_added <= (sc->max_vlans + 1))
2229 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2231 POCE_SOFTC sc = ifp->if_softc;
2233 if (ifp->if_softc != arg)
2235 if ((vtag == 0) || (vtag > 4095))
2238 sc->vlan_tag[vtag] = 0;
2245 * A max of 64 vlans can be configured in BE. If the user configures
2246 * more, place the card in vlan promiscuous mode.
2249 oce_vid_config(POCE_SOFTC sc)
2251 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2252 uint16_t ntags = 0, i;
2255 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2256 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2257 for (i = 0; i < MAX_VLANS; i++) {
2258 if (sc->vlan_tag[i]) {
2259 vtags[ntags].vtag = i;
2264 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2265 vtags, ntags, 1, 0);
2267 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2274 oce_mac_addr_set(POCE_SOFTC sc)
2276 uint32_t old_pmac_id = sc->pmac_id;
2280 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2281 sc->macaddr.size_of_struct);
2285 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2286 sc->if_id, &sc->pmac_id);
2288 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2289 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2290 sc->macaddr.size_of_struct);
2293 device_printf(sc->dev, "Failed update macaddress\n");
2299 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2301 POCE_SOFTC sc = ifp->if_softc;
2302 struct ifreq *ifr = (struct ifreq *)data;
2304 char cookie[32] = {0};
2305 void *priv_data = ifr_data_get_ptr(ifr);
2309 OCE_DMA_MEM dma_mem;
2310 struct mbx_common_get_cntl_attr *fw_cmd;
2312 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2315 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2318 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2319 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2322 req_size = le32toh(req.u0.req.request_length);
2323 if (req_size > 65536)
2326 req_size += sizeof(struct mbx_hdr);
2327 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2331 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2336 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2342 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2346 firmware is filling all the attributes for this ioctl except
2347 the driver version..so fill it
2349 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2350 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2351 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2352 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2356 oce_dma_free(sc, &dma_mem);
2362 oce_eqd_set_periodic(POCE_SOFTC sc)
2364 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2365 struct oce_aic_obj *aic;
2367 uint64_t now = 0, delta;
2368 int eqd, i, num = 0;
2369 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2373 #define ticks_to_msecs(t) (1000 * (t) / hz)
2375 for (i = 0 ; i < sc->neqs; i++) {
2377 aic = &sc->aic_obj[i];
2378 /* When setting the static eq delay from the user space */
2388 rxpkts = rq->rx_stats.rx_pkts;
2391 if (i + 1 < sc->nrqs) {
2393 rxpkts += rq->rx_stats.rx_pkts;
2397 tx_reqs = wq->tx_stats.tx_reqs;
2402 if (!aic->ticks || now < aic->ticks ||
2403 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2404 aic->prev_rxpkts = rxpkts;
2405 aic->prev_txreqs = tx_reqs;
2410 delta = ticks_to_msecs(now - aic->ticks);
2412 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2413 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2414 eqd = (pps / 15000) << 2;
2418 /* Make sure that the eq delay is in the known range */
2419 eqd = min(eqd, aic->max_eqd);
2420 eqd = max(eqd, aic->min_eqd);
2422 aic->prev_rxpkts = rxpkts;
2423 aic->prev_txreqs = tx_reqs;
2427 if (eqd != aic->cur_eqd) {
2428 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2429 set_eqd[num].eq_id = eqo->eq_id;
2435 /* Is there atleast one eq that needs to be modified? */
2436 for(i = 0; i < num; i += 8) {
2438 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2440 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2445 static void oce_detect_hw_error(POCE_SOFTC sc)
2448 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2449 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2456 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2457 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2458 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2459 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2462 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2463 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2464 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2465 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2467 ue_low = (ue_low & ~ue_low_mask);
2468 ue_high = (ue_high & ~ue_high_mask);
2471 /* On certain platforms BE hardware can indicate spurious UEs.
2472 * Allow the h/w to stop working completely in case of a real UE.
2473 * Hence not setting the hw_error for UE detection.
2475 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2476 sc->hw_error = TRUE;
2477 device_printf(sc->dev, "Error detected in the card\n");
2480 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2481 device_printf(sc->dev,
2482 "ERR: sliport status 0x%x\n", sliport_status);
2483 device_printf(sc->dev,
2484 "ERR: sliport error1 0x%x\n", sliport_err1);
2485 device_printf(sc->dev,
2486 "ERR: sliport error2 0x%x\n", sliport_err2);
2490 for (i = 0; ue_low; ue_low >>= 1, i++) {
2492 device_printf(sc->dev, "UE: %s bit set\n",
2493 ue_status_low_desc[i]);
2498 for (i = 0; ue_high; ue_high >>= 1, i++) {
2500 device_printf(sc->dev, "UE: %s bit set\n",
2501 ue_status_hi_desc[i]);
2509 oce_local_timer(void *arg)
2511 POCE_SOFTC sc = arg;
2514 oce_detect_hw_error(sc);
2515 oce_refresh_nic_stats(sc);
2516 oce_refresh_queue_stats(sc);
2517 oce_mac_addr_set(sc);
2520 for (i = 0; i < sc->nwqs; i++)
2521 oce_tx_restart(sc, sc->wq[i]);
2523 /* calculate and set the eq delay for optimal interrupt rate */
2524 if (IS_BE(sc) || IS_SH(sc))
2525 oce_eqd_set_periodic(sc);
2527 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2531 oce_tx_compl_clean(POCE_SOFTC sc)
2534 int i = 0, timeo = 0, num_wqes = 0;
2535 int pending_txqs = sc->nwqs;
2537 /* Stop polling for compls when HW has been silent for 10ms or
2538 * hw_error or no outstanding completions expected
2541 pending_txqs = sc->nwqs;
2543 for_all_wq_queues(sc, wq, i) {
2544 num_wqes = oce_wq_handler(wq);
2549 if(!wq->ring->num_used)
2553 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2559 for_all_wq_queues(sc, wq, i) {
2560 while(wq->ring->num_used) {
2561 LOCK(&wq->tx_compl_lock);
2562 oce_process_tx_completion(wq);
2563 UNLOCK(&wq->tx_compl_lock);
2569 /* NOTE : This should only be called holding
2573 oce_if_deactivate(POCE_SOFTC sc)
2580 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2582 oce_tx_compl_clean(sc);
2584 /* Stop intrs and finish any bottom halves pending */
2585 oce_hw_intr_disable(sc);
2587 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2588 any other lock. So unlock device lock and require after
2589 completing taskqueue_drain.
2591 UNLOCK(&sc->dev_lock);
2592 for (i = 0; i < sc->intr_count; i++) {
2593 if (sc->intrs[i].tq != NULL) {
2594 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2597 LOCK(&sc->dev_lock);
2599 /* Delete RX queue in card with flush param */
2602 /* Invalidate any pending cq and eq entries*/
2603 for_all_evnt_queues(sc, eq, i)
2605 for_all_rq_queues(sc, rq, i)
2606 oce_drain_rq_cq(rq);
2607 for_all_wq_queues(sc, wq, i)
2608 oce_drain_wq_cq(wq);
2610 /* But still we need to get MCC aync events.
2611 So enable intrs and also arm first EQ
2613 oce_hw_intr_enable(sc);
2614 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2621 oce_if_activate(POCE_SOFTC sc)
2628 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2630 oce_hw_intr_disable(sc);
2634 for_all_rq_queues(sc, rq, i) {
2635 rc = oce_start_rq(rq);
2637 device_printf(sc->dev, "Unable to start RX\n");
2640 for_all_wq_queues(sc, wq, i) {
2641 rc = oce_start_wq(wq);
2643 device_printf(sc->dev, "Unable to start TX\n");
2647 for_all_evnt_queues(sc, eq, i)
2648 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2650 oce_hw_intr_enable(sc);
2655 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2657 /* Update Link status */
2658 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2659 ASYNC_EVENT_LINK_UP) {
2660 sc->link_status = ASYNC_EVENT_LINK_UP;
2661 if_link_state_change(sc->ifp, LINK_STATE_UP);
2663 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2664 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2669 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2670 struct oce_async_evt_grp5_os2bmc *evt)
2672 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2673 if (evt->u.s.mgmt_enable)
2674 sc->flags |= OCE_FLAGS_OS2BMC;
2678 sc->bmc_filt_mask = evt->u.s.arp_filter;
2679 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2680 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2681 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2682 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2683 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2684 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2685 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2686 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2690 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2692 struct oce_async_event_grp5_pvid_state *gcqe;
2693 struct oce_async_evt_grp5_os2bmc *bmccqe;
2695 switch (cqe->u0.s.async_type) {
2696 case ASYNC_EVENT_PVID_STATE:
2698 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2700 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2704 case ASYNC_EVENT_OS2BMC:
2705 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2706 oce_async_grp5_osbmc_process(sc, bmccqe);
2713 /* Handle the Completion Queue for the Mailbox/Async notifications */
2715 oce_mq_handler(void *arg)
2717 struct oce_mq *mq = (struct oce_mq *)arg;
2718 POCE_SOFTC sc = mq->parent;
2719 struct oce_cq *cq = mq->cq;
2720 int num_cqes = 0, evt_type = 0, optype = 0;
2721 struct oce_mq_cqe *cqe;
2722 struct oce_async_cqe_link_state *acqe;
2723 struct oce_async_event_qnq *dbgcqe;
2726 bus_dmamap_sync(cq->ring->dma.tag,
2727 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2728 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2730 while (cqe->u0.dw[3]) {
2731 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2732 if (cqe->u0.s.async_event) {
2733 evt_type = cqe->u0.s.event_type;
2734 optype = cqe->u0.s.async_type;
2735 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2736 /* Link status evt */
2737 acqe = (struct oce_async_cqe_link_state *)cqe;
2738 process_link_state(sc, acqe);
2739 } else if (evt_type == ASYNC_EVENT_GRP5) {
2740 oce_process_grp5_events(sc, cqe);
2741 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2742 optype == ASYNC_EVENT_DEBUG_QNQ) {
2743 dbgcqe = (struct oce_async_event_qnq *)cqe;
2745 sc->qnqid = dbgcqe->vlan_tag;
2746 sc->qnq_debug_event = TRUE;
2750 RING_GET(cq->ring, 1);
2751 bus_dmamap_sync(cq->ring->dma.tag,
2752 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2753 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2758 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2765 setup_max_queues_want(POCE_SOFTC sc)
2767 /* Check if it is FLEX machine. Is so dont use RSS */
2768 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2769 (sc->function_mode & FNM_UMC_MODE) ||
2770 (sc->function_mode & FNM_VNIC_MODE) ||
2771 (!is_rss_enabled(sc)) ||
2776 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2777 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2780 if (IS_BE2(sc) && is_rss_enabled(sc))
2781 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2786 update_queues_got(POCE_SOFTC sc)
2788 if (is_rss_enabled(sc)) {
2789 sc->nrqs = sc->intr_count + 1;
2790 sc->nwqs = sc->intr_count;
2801 oce_check_ipv6_ext_hdr(struct mbuf *m)
2803 struct ether_header *eh = mtod(m, struct ether_header *);
2804 caddr_t m_datatemp = m->m_data;
2806 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2807 m->m_data += sizeof(struct ether_header);
2808 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2810 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2811 (ip6->ip6_nxt != IPPROTO_UDP)){
2812 struct ip6_ext *ip6e = NULL;
2813 m->m_data += sizeof(struct ip6_hdr);
2815 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2816 if(ip6e->ip6e_len == 0xff) {
2817 m->m_data = m_datatemp;
2821 m->m_data = m_datatemp;
2827 is_be3_a1(POCE_SOFTC sc)
2829 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2835 static struct mbuf *
2836 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2838 uint16_t vlan_tag = 0;
2843 /* Embed vlan tag in the packet if it is not part of it */
2844 if(m->m_flags & M_VLANTAG) {
2845 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2846 m->m_flags &= ~M_VLANTAG;
2849 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2852 vlan_tag = sc->pvid;
2858 m = ether_vlanencap(m, vlan_tag);
2862 m = ether_vlanencap(m, sc->qnqid);
2871 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2873 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2874 oce_check_ipv6_ext_hdr(m)) {
2881 oce_get_config(POCE_SOFTC sc)
2884 uint32_t max_rss = 0;
2886 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2887 max_rss = OCE_LEGACY_MODE_RSS;
2889 max_rss = OCE_MAX_RSS;
2892 rc = oce_get_profile_config(sc, max_rss);
2894 sc->nwqs = OCE_MAX_WQ;
2895 sc->nrssqs = max_rss;
2896 sc->nrqs = sc->nrssqs + 1;
2899 else { /* For BE3 don't rely on fw for determining the resources */
2900 sc->nrssqs = max_rss;
2901 sc->nrqs = sc->nrssqs + 1;
2902 sc->nwqs = OCE_MAX_WQ;
2903 sc->max_vlans = MAX_VLANFILTER_SIZE;
2908 oce_rdma_close(void)
2910 if (oce_rdma_if != NULL) {
2916 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2918 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2922 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2925 struct oce_dev_info di;
2928 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2932 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2933 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2937 rdma_info->close = oce_rdma_close;
2938 rdma_info->mbox_post = oce_mbox_post;
2939 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2940 rdma_info->get_mac_addr = oce_get_mac_addr;
2942 oce_rdma_if = rdma_if;
2945 while (sc != NULL) {
2946 if (oce_rdma_if->announce != NULL) {
2947 memset(&di, 0, sizeof(di));
2951 di.db_bhandle = sc->db_bhandle;
2952 di.db_btag = sc->db_btag;
2953 di.db_page_size = 4096;
2954 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2955 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2956 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2957 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2959 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2961 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2962 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2963 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2964 di.msix.start_vector = sc->intr_count;
2965 for (i=0; i<di.msix.num_vectors; i++) {
2966 di.msix.vector_list[i] = sc->intrs[i].vector;
2970 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2971 di.vendor_id = pci_get_vendor(sc->dev);
2972 di.dev_id = pci_get_device(sc->dev);
2974 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2975 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2978 rdma_if->announce(&di);
2987 oce_read_env_variables( POCE_SOFTC sc )
2992 /* read if user wants to enable hwlro or swlro */
2993 //value = getenv("oce_enable_hwlro");
2994 if(value && IS_SH(sc)) {
2995 sc->enable_hwlro = strtol(value, NULL, 10);
2996 if(sc->enable_hwlro) {
2997 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
2999 device_printf(sc->dev, "no hardware lro support\n");
3000 device_printf(sc->dev, "software lro enabled\n");
3001 sc->enable_hwlro = 0;
3003 device_printf(sc->dev, "hardware lro enabled\n");
3004 oce_max_rsp_handled = 32;
3007 device_printf(sc->dev, "software lro enabled\n");
3010 sc->enable_hwlro = 0;
3013 /* read mbuf size */
3014 //value = getenv("oce_rq_buf_size");
3015 if(value && IS_SH(sc)) {
3016 oce_rq_buf_size = strtol(value, NULL, 10);
3017 switch(oce_rq_buf_size) {
3025 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
3026 oce_rq_buf_size = 2048;