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 &&
597 i2c.dev_addr != PAGE_NUM_A2) {
602 if (i2c.len > sizeof(i2c.data)) {
607 rc = oce_mbox_read_transrecv_data(sc, i2c.dev_addr);
613 if (i2c.dev_addr == PAGE_NUM_A0)
616 offset = TRANSCEIVER_A0_SIZE + i2c.offset;
618 memcpy(&i2c.data[0], &sfp_vpd_dump_buffer[offset], i2c.len);
620 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
624 rc = priv_check(curthread, PRIV_DRIVER);
627 rc = oce_handle_passthrough(ifp, data);
630 rc = ether_ioctl(ifp, command, data);
645 if (sc->ifp->if_flags & IFF_UP) {
646 oce_if_deactivate(sc);
650 UNLOCK(&sc->dev_lock);
656 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
658 POCE_SOFTC sc = ifp->if_softc;
659 struct oce_wq *wq = NULL;
663 if (!sc->link_status)
666 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
667 queue_index = m->m_pkthdr.flowid % sc->nwqs;
669 wq = sc->wq[queue_index];
672 status = oce_multiq_transmit(ifp, m, wq);
673 UNLOCK(&wq->tx_lock);
681 oce_multiq_flush(struct ifnet *ifp)
683 POCE_SOFTC sc = ifp->if_softc;
687 for (i = 0; i < sc->nwqs; i++) {
688 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
696 /*****************************************************************************
697 * Driver interrupt routines functions *
698 *****************************************************************************/
701 oce_intr(void *arg, int pending)
704 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
705 POCE_SOFTC sc = ii->sc;
706 struct oce_eq *eq = ii->eq;
708 struct oce_cq *cq = NULL;
712 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
713 BUS_DMASYNC_POSTWRITE);
715 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
719 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
720 BUS_DMASYNC_POSTWRITE);
721 RING_GET(eq->ring, 1);
727 goto eq_arm; /* Spurious */
729 /* Clear EQ entries, but dont arm */
730 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
732 /* Process TX, RX and MCC. But dont arm CQ*/
733 for (i = 0; i < eq->cq_valid; i++) {
735 (*cq->cq_handler)(cq->cb_arg);
738 /* Arm all cqs connected to this EQ */
739 for (i = 0; i < eq->cq_valid; i++) {
741 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
745 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
752 oce_setup_intr(POCE_SOFTC sc)
754 int rc = 0, use_intx = 0;
755 int vector = 0, req_vectors = 0;
756 int tot_req_vectors, tot_vectors;
758 if (is_rss_enabled(sc))
759 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
763 tot_req_vectors = req_vectors;
764 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
765 if (req_vectors > 1) {
766 tot_req_vectors += OCE_RDMA_VECTORS;
767 sc->roce_intr_count = OCE_RDMA_VECTORS;
771 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
772 sc->intr_count = req_vectors;
773 tot_vectors = tot_req_vectors;
774 rc = pci_alloc_msix(sc->dev, &tot_vectors);
777 pci_release_msi(sc->dev);
779 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
780 if (tot_vectors < tot_req_vectors) {
781 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
782 sc->roce_intr_count = (tot_vectors / 2);
784 sc->intr_count = tot_vectors - sc->roce_intr_count;
787 sc->intr_count = tot_vectors;
789 sc->flags |= OCE_FLAGS_USING_MSIX;
797 /* Scale number of queues based on intr we got */
798 update_queues_got(sc);
801 device_printf(sc->dev, "Using legacy interrupt\n");
802 rc = oce_alloc_intr(sc, vector, oce_intr);
806 for (; vector < sc->intr_count; vector++) {
807 rc = oce_alloc_intr(sc, vector, oce_intr);
821 oce_fast_isr(void *arg)
823 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
824 POCE_SOFTC sc = ii->sc;
829 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
831 taskqueue_enqueue(ii->tq, &ii->task);
835 return FILTER_HANDLED;
840 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
845 if (vector >= OCE_MAX_EQ)
848 ii = &sc->intrs[vector];
850 /* Set the resource id for the interrupt.
851 * MSIx is vector + 1 for the resource id,
852 * INTx is 0 for the resource id.
854 if (sc->flags & OCE_FLAGS_USING_MSIX)
858 ii->intr_res = bus_alloc_resource_any(sc->dev,
860 &rr, RF_ACTIVE|RF_SHAREABLE);
862 if (ii->intr_res == NULL) {
863 device_printf(sc->dev,
864 "Could not allocate interrupt\n");
869 TASK_INIT(&ii->task, 0, isr, ii);
871 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
872 ii->tq = taskqueue_create_fast(ii->task_name,
874 taskqueue_thread_enqueue,
876 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
877 device_get_nameunit(sc->dev));
880 rc = bus_setup_intr(sc->dev,
883 oce_fast_isr, NULL, ii, &ii->tag);
890 oce_intr_free(POCE_SOFTC sc)
894 for (i = 0; i < sc->intr_count; i++) {
896 if (sc->intrs[i].tag != NULL)
897 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
899 if (sc->intrs[i].tq != NULL)
900 taskqueue_free(sc->intrs[i].tq);
902 if (sc->intrs[i].intr_res != NULL)
903 bus_release_resource(sc->dev, SYS_RES_IRQ,
905 sc->intrs[i].intr_res);
906 sc->intrs[i].tag = NULL;
907 sc->intrs[i].intr_res = NULL;
910 if (sc->flags & OCE_FLAGS_USING_MSIX)
911 pci_release_msi(sc->dev);
917 /******************************************************************************
918 * Media callbacks functions *
919 ******************************************************************************/
922 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
924 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
927 req->ifm_status = IFM_AVALID;
928 req->ifm_active = IFM_ETHER;
930 if (sc->link_status == 1)
931 req->ifm_status |= IFM_ACTIVE;
935 switch (sc->link_speed) {
936 case 1: /* 10 Mbps */
937 req->ifm_active |= IFM_10_T | IFM_FDX;
940 case 2: /* 100 Mbps */
941 req->ifm_active |= IFM_100_TX | IFM_FDX;
945 req->ifm_active |= IFM_1000_T | IFM_FDX;
948 case 4: /* 10 Gbps */
949 req->ifm_active |= IFM_10G_SR | IFM_FDX;
952 case 5: /* 20 Gbps */
953 req->ifm_active |= IFM_10G_SR | IFM_FDX;
956 case 6: /* 25 Gbps */
957 req->ifm_active |= IFM_10G_SR | IFM_FDX;
960 case 7: /* 40 Gbps */
961 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
974 oce_media_change(struct ifnet *ifp)
980 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
981 struct mbuf *m, boolean_t *os2bmc,
984 struct ether_header *eh = NULL;
986 eh = mtod(m, struct ether_header *);
988 if (!is_os2bmc_enabled(sc) || *os2bmc) {
992 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
995 if (is_mc_allowed_on_bmc(sc, eh) ||
996 is_bc_allowed_on_bmc(sc, eh) ||
997 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
1002 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
1003 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1004 uint8_t nexthdr = ip6->ip6_nxt;
1005 if (nexthdr == IPPROTO_ICMPV6) {
1006 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
1007 switch (icmp6->icmp6_type) {
1008 case ND_ROUTER_ADVERT:
1009 *os2bmc = is_ipv6_ra_filt_enabled(sc);
1011 case ND_NEIGHBOR_ADVERT:
1012 *os2bmc = is_ipv6_na_filt_enabled(sc);
1020 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
1021 struct ip *ip = mtod(m, struct ip *);
1022 int iphlen = ip->ip_hl << 2;
1023 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
1024 switch (uh->uh_dport) {
1025 case DHCP_CLIENT_PORT:
1026 *os2bmc = is_dhcp_client_filt_enabled(sc);
1028 case DHCP_SERVER_PORT:
1029 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
1031 case NET_BIOS_PORT1:
1032 case NET_BIOS_PORT2:
1033 *os2bmc = is_nbios_filt_enabled(sc);
1035 case DHCPV6_RAS_PORT:
1036 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1044 *m_new = m_dup(m, M_NOWAIT);
1049 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1055 /*****************************************************************************
1056 * Transmit routines functions *
1057 *****************************************************************************/
1060 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1062 int rc = 0, i, retry_cnt = 0;
1063 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1064 struct mbuf *m, *m_temp, *m_new = NULL;
1065 struct oce_wq *wq = sc->wq[wq_index];
1066 struct oce_packet_desc *pd;
1067 struct oce_nic_hdr_wqe *nichdr;
1068 struct oce_nic_frag_wqe *nicfrag;
1069 struct ether_header *eh = NULL;
1072 boolean_t complete = TRUE;
1073 boolean_t os2bmc = FALSE;
1079 if (!(m->m_flags & M_PKTHDR)) {
1084 /* Don't allow non-TSO packets longer than MTU */
1085 if (!is_tso_pkt(m)) {
1086 eh = mtod(m, struct ether_header *);
1087 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1091 if(oce_tx_asic_stall_verify(sc, m)) {
1092 m = oce_insert_vlan_tag(sc, m, &complete);
1094 device_printf(sc->dev, "Insertion unsuccessful\n");
1100 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1101 * may cause a transmit stall on that port. So the work-around is to
1102 * pad short packets (<= 32 bytes) to a 36-byte length.
1104 if(IS_SH(sc) || IS_XE201(sc) ) {
1105 if(m->m_pkthdr.len <= 32) {
1107 bzero((void *)buf, 36);
1108 m_append(m, (36 - m->m_pkthdr.len), buf);
1113 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1114 /* consolidate packet buffers for TSO/LSO segment offload */
1115 #if defined(INET6) || defined(INET)
1116 m = oce_tso_setup(sc, mpp);
1127 pd = &wq->pckts[wq->pkt_desc_head];
1130 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1132 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1134 num_wqes = pd->nsegs + 1;
1135 if (IS_BE(sc) || IS_SH(sc)) {
1136 /*Dummy required only for BE3.*/
1140 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1141 bus_dmamap_unload(wq->tag, pd->map);
1144 atomic_store_rel_int(&wq->pkt_desc_head,
1145 (wq->pkt_desc_head + 1) % \
1146 OCE_WQ_PACKET_ARRAY_SIZE);
1147 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1151 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1152 nichdr->u0.dw[0] = 0;
1153 nichdr->u0.dw[1] = 0;
1154 nichdr->u0.dw[2] = 0;
1155 nichdr->u0.dw[3] = 0;
1157 nichdr->u0.s.complete = complete;
1158 nichdr->u0.s.mgmt = os2bmc;
1159 nichdr->u0.s.event = 1;
1160 nichdr->u0.s.crc = 1;
1161 nichdr->u0.s.forward = 0;
1162 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1163 nichdr->u0.s.udpcs =
1164 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1165 nichdr->u0.s.tcpcs =
1166 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1167 nichdr->u0.s.num_wqe = num_wqes;
1168 nichdr->u0.s.total_length = m->m_pkthdr.len;
1170 if (m->m_flags & M_VLANTAG) {
1171 nichdr->u0.s.vlan = 1; /*Vlan present*/
1172 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1175 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1176 if (m->m_pkthdr.tso_segsz) {
1177 nichdr->u0.s.lso = 1;
1178 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1180 if (!IS_BE(sc) || !IS_SH(sc))
1181 nichdr->u0.s.ipcs = 1;
1184 RING_PUT(wq->ring, 1);
1185 atomic_add_int(&wq->ring->num_used, 1);
1187 for (i = 0; i < pd->nsegs; i++) {
1189 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1190 struct oce_nic_frag_wqe);
1191 nicfrag->u0.s.rsvd0 = 0;
1192 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1193 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1194 nicfrag->u0.s.frag_len = segs[i].ds_len;
1195 pd->wqe_idx = wq->ring->pidx;
1196 RING_PUT(wq->ring, 1);
1197 atomic_add_int(&wq->ring->num_used, 1);
1199 if (num_wqes > (pd->nsegs + 1)) {
1201 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1202 struct oce_nic_frag_wqe);
1203 nicfrag->u0.dw[0] = 0;
1204 nicfrag->u0.dw[1] = 0;
1205 nicfrag->u0.dw[2] = 0;
1206 nicfrag->u0.dw[3] = 0;
1207 pd->wqe_idx = wq->ring->pidx;
1208 RING_PUT(wq->ring, 1);
1209 atomic_add_int(&wq->ring->num_used, 1);
1213 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1214 wq->tx_stats.tx_reqs++;
1215 wq->tx_stats.tx_wrbs += num_wqes;
1216 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1217 wq->tx_stats.tx_pkts++;
1219 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1220 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1221 reg_value = (num_wqes << 16) | wq->wq_id;
1223 /* if os2bmc is not enabled or if the pkt is already tagged as
1226 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1228 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1230 } else if (rc == EFBIG) {
1231 if (retry_cnt == 0) {
1232 m_temp = m_defrag(m, M_NOWAIT);
1237 retry_cnt = retry_cnt + 1;
1241 } else if (rc == ENOMEM)
1261 oce_process_tx_completion(struct oce_wq *wq)
1263 struct oce_packet_desc *pd;
1264 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1267 pd = &wq->pckts[wq->pkt_desc_tail];
1268 atomic_store_rel_int(&wq->pkt_desc_tail,
1269 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1270 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1271 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1272 bus_dmamap_unload(wq->tag, pd->map);
1279 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1280 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1281 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1282 oce_tx_restart(sc, wq);
1289 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1292 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1295 #if __FreeBSD_version >= 800000
1296 if (!drbr_empty(sc->ifp, wq->br))
1298 if (!IFQ_DRV_IS_EMPTY(&sc->ifp->if_snd))
1300 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1305 #if defined(INET6) || defined(INET)
1306 static struct mbuf *
1307 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1314 struct ip6_hdr *ip6;
1316 struct ether_vlan_header *eh;
1319 int total_len = 0, ehdrlen = 0;
1323 if (M_WRITABLE(m) == 0) {
1324 m = m_dup(*mpp, M_NOWAIT);
1331 eh = mtod(m, struct ether_vlan_header *);
1332 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1333 etype = ntohs(eh->evl_proto);
1334 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1336 etype = ntohs(eh->evl_encap_proto);
1337 ehdrlen = ETHER_HDR_LEN;
1343 ip = (struct ip *)(m->m_data + ehdrlen);
1344 if (ip->ip_p != IPPROTO_TCP)
1346 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1348 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1352 case ETHERTYPE_IPV6:
1353 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1354 if (ip6->ip6_nxt != IPPROTO_TCP)
1356 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1358 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1365 m = m_pullup(m, total_len);
1372 #endif /* INET6 || INET */
1375 oce_tx_task(void *arg, int npending)
1377 struct oce_wq *wq = arg;
1378 POCE_SOFTC sc = wq->parent;
1379 struct ifnet *ifp = sc->ifp;
1382 #if __FreeBSD_version >= 800000
1384 rc = oce_multiq_transmit(ifp, NULL, wq);
1386 device_printf(sc->dev,
1387 "TX[%d] restart failed\n", wq->queue_index);
1389 UNLOCK(&wq->tx_lock);
1398 oce_start(struct ifnet *ifp)
1400 POCE_SOFTC sc = ifp->if_softc;
1403 int def_q = 0; /* Defualt tx queue is 0*/
1405 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1409 if (!sc->link_status)
1413 IF_DEQUEUE(&sc->ifp->if_snd, m);
1417 LOCK(&sc->wq[def_q]->tx_lock);
1418 rc = oce_tx(sc, &m, def_q);
1419 UNLOCK(&sc->wq[def_q]->tx_lock);
1422 sc->wq[def_q]->tx_stats.tx_stops ++;
1423 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1424 IFQ_DRV_PREPEND(&ifp->if_snd, m);
1430 ETHER_BPF_MTAP(ifp, m);
1438 /* Handle the Completion Queue for transmit */
1440 oce_wq_handler(void *arg)
1442 struct oce_wq *wq = (struct oce_wq *)arg;
1443 POCE_SOFTC sc = wq->parent;
1444 struct oce_cq *cq = wq->cq;
1445 struct oce_nic_tx_cqe *cqe;
1448 LOCK(&wq->tx_compl_lock);
1449 bus_dmamap_sync(cq->ring->dma.tag,
1450 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1451 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1452 while (cqe->u0.dw[3]) {
1453 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe));
1455 wq->ring->cidx = cqe->u0.s.wqe_index + 1;
1456 if (wq->ring->cidx >= wq->ring->num_items)
1457 wq->ring->cidx -= wq->ring->num_items;
1459 oce_process_tx_completion(wq);
1460 wq->tx_stats.tx_compl++;
1462 RING_GET(cq->ring, 1);
1463 bus_dmamap_sync(cq->ring->dma.tag,
1464 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1466 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1471 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
1473 UNLOCK(&wq->tx_compl_lock);
1479 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq)
1481 POCE_SOFTC sc = ifp->if_softc;
1482 int status = 0, queue_index = 0;
1483 struct mbuf *next = NULL;
1484 struct buf_ring *br = NULL;
1487 queue_index = wq->queue_index;
1489 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1492 status = drbr_enqueue(ifp, br, m);
1497 if ((status = drbr_enqueue(ifp, br, m)) != 0)
1500 while ((next = drbr_peek(ifp, br)) != NULL) {
1501 if (oce_tx(sc, &next, queue_index)) {
1503 drbr_advance(ifp, br);
1505 drbr_putback(ifp, br, next);
1506 wq->tx_stats.tx_stops ++;
1507 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1511 drbr_advance(ifp, br);
1512 if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
1513 if (next->m_flags & M_MCAST)
1514 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1515 ETHER_BPF_MTAP(ifp, next);
1524 /*****************************************************************************
1525 * Receive routines functions *
1526 *****************************************************************************/
1529 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2)
1532 struct ether_header *eh = NULL;
1533 struct tcphdr *tcp_hdr = NULL;
1534 struct ip *ip4_hdr = NULL;
1535 struct ip6_hdr *ip6 = NULL;
1536 uint32_t payload_len = 0;
1538 eh = mtod(m, struct ether_header *);
1539 /* correct IP header */
1540 if(!cqe2->ipv6_frame) {
1541 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header));
1542 ip4_hdr->ip_ttl = cqe2->frame_lifespan;
1543 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header));
1544 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip));
1546 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header));
1547 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan;
1548 payload_len = cqe2->coalesced_size - sizeof(struct ether_header)
1549 - sizeof(struct ip6_hdr);
1550 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len);
1551 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr));
1554 /* correct tcp header */
1555 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num);
1557 tcp_hdr->th_flags |= TH_PUSH;
1559 tcp_hdr->th_win = htons(cqe2->tcp_window);
1560 tcp_hdr->th_sum = 0xffff;
1562 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2);
1563 *p = cqe1->tcp_timestamp_val;
1564 *(p+1) = cqe1->tcp_timestamp_ecr;
1571 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m)
1573 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1574 uint32_t i = 0, frag_len = 0;
1575 uint32_t len = cqe_info->pkt_size;
1576 struct oce_packet_desc *pd;
1577 struct mbuf *tail = NULL;
1579 for (i = 0; i < cqe_info->num_frags; i++) {
1580 if (rq->ring->cidx == rq->ring->pidx) {
1581 device_printf(sc->dev,
1582 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n");
1585 pd = &rq->pckts[rq->ring->cidx];
1587 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1588 bus_dmamap_unload(rq->tag, pd->map);
1589 RING_GET(rq->ring, 1);
1592 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len;
1593 pd->mbuf->m_len = frag_len;
1596 /* additional fragments */
1597 pd->mbuf->m_flags &= ~M_PKTHDR;
1598 tail->m_next = pd->mbuf;
1600 tail->m_nextpkt = NULL;
1603 /* first fragment, fill out much of the packet header */
1604 pd->mbuf->m_pkthdr.len = len;
1606 pd->mbuf->m_nextpkt = NULL;
1607 pd->mbuf->m_pkthdr.csum_flags = 0;
1608 if (IF_CSUM_ENABLED(sc)) {
1609 if (cqe_info->l4_cksum_pass) {
1610 if(!cqe_info->ipv6_frame) { /* IPV4 */
1611 pd->mbuf->m_pkthdr.csum_flags |=
1612 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1613 }else { /* IPV6 frame */
1615 pd->mbuf->m_pkthdr.csum_flags |=
1616 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1619 pd->mbuf->m_pkthdr.csum_data = 0xffff;
1621 if (cqe_info->ip_cksum_pass) {
1622 pd->mbuf->m_pkthdr.csum_flags |=
1623 (CSUM_IP_CHECKED|CSUM_IP_VALID);
1626 *m = tail = pd->mbuf;
1636 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2)
1638 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1639 struct nic_hwlro_cqe_part1 *cqe1 = NULL;
1640 struct mbuf *m = NULL;
1641 struct oce_common_cqe_info cq_info;
1645 cq_info.pkt_size = cqe->pkt_size;
1646 cq_info.vtag = cqe->vlan_tag;
1647 cq_info.l4_cksum_pass = cqe->l4_cksum_pass;
1648 cq_info.ip_cksum_pass = cqe->ip_cksum_pass;
1649 cq_info.ipv6_frame = cqe->ipv6_frame;
1650 cq_info.vtp = cqe->vtp;
1651 cq_info.qnq = cqe->qnq;
1653 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe;
1654 cq_info.pkt_size = cqe2->coalesced_size;
1655 cq_info.vtag = cqe2->vlan_tag;
1656 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass;
1657 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass;
1658 cq_info.ipv6_frame = cqe2->ipv6_frame;
1659 cq_info.vtp = cqe2->vtp;
1660 cq_info.qnq = cqe1->qnq;
1663 cq_info.vtag = BSWAP_16(cq_info.vtag);
1665 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size;
1666 if(cq_info.pkt_size % rq->cfg.frag_size)
1667 cq_info.num_frags++;
1669 oce_rx_mbuf_chain(rq, &cq_info, &m);
1673 //assert(cqe2->valid != 0);
1675 //assert(cqe2->cqe_type != 2);
1676 oce_correct_header(m, cqe1, cqe2);
1679 m->m_pkthdr.rcvif = sc->ifp;
1680 #if __FreeBSD_version >= 800000
1681 if (rq->queue_index)
1682 m->m_pkthdr.flowid = (rq->queue_index - 1);
1684 m->m_pkthdr.flowid = rq->queue_index;
1685 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1687 /* This deternies if vlan tag is Valid */
1689 if (sc->function_mode & FNM_FLEX10_MODE) {
1690 /* FLEX10. If QnQ is not set, neglect VLAN */
1692 m->m_pkthdr.ether_vtag = cq_info.vtag;
1693 m->m_flags |= M_VLANTAG;
1695 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1696 /* In UMC mode generally pvid will be striped by
1697 hw. But in some cases we have seen it comes
1698 with pvid. So if pvid == vlan, neglect vlan.
1700 m->m_pkthdr.ether_vtag = cq_info.vtag;
1701 m->m_flags |= M_VLANTAG;
1704 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1706 (*sc->ifp->if_input) (sc->ifp, m);
1708 /* Update rx stats per queue */
1709 rq->rx_stats.rx_pkts++;
1710 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1711 rq->rx_stats.rx_frags += cq_info.num_frags;
1712 rq->rx_stats.rx_ucast_pkts++;
1718 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1720 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1722 struct mbuf *m = NULL;
1723 struct oce_common_cqe_info cq_info;
1726 /* Is it a flush compl that has no data */
1727 if(!cqe->u0.s.num_fragments)
1730 len = cqe->u0.s.pkt_size;
1732 /*partial DMA workaround for Lancer*/
1733 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1737 if (!oce_cqe_portid_valid(sc, cqe)) {
1738 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1742 /* Get vlan_tag value */
1743 if(IS_BE(sc) || IS_SH(sc))
1744 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1746 vtag = cqe->u0.s.vlan_tag;
1748 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1749 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1750 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1751 cq_info.num_frags = cqe->u0.s.num_fragments;
1752 cq_info.pkt_size = cqe->u0.s.pkt_size;
1754 oce_rx_mbuf_chain(rq, &cq_info, &m);
1757 m->m_pkthdr.rcvif = sc->ifp;
1758 #if __FreeBSD_version >= 800000
1759 if (rq->queue_index)
1760 m->m_pkthdr.flowid = (rq->queue_index - 1);
1762 m->m_pkthdr.flowid = rq->queue_index;
1763 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1765 /* This deternies if vlan tag is Valid */
1766 if (oce_cqe_vtp_valid(sc, cqe)) {
1767 if (sc->function_mode & FNM_FLEX10_MODE) {
1768 /* FLEX10. If QnQ is not set, neglect VLAN */
1769 if (cqe->u0.s.qnq) {
1770 m->m_pkthdr.ether_vtag = vtag;
1771 m->m_flags |= M_VLANTAG;
1773 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1774 /* In UMC mode generally pvid will be striped by
1775 hw. But in some cases we have seen it comes
1776 with pvid. So if pvid == vlan, neglect vlan.
1778 m->m_pkthdr.ether_vtag = vtag;
1779 m->m_flags |= M_VLANTAG;
1783 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1784 #if defined(INET6) || defined(INET)
1785 /* Try to queue to LRO */
1786 if (IF_LRO_ENABLED(sc) &&
1787 (cqe->u0.s.ip_cksum_pass) &&
1788 (cqe->u0.s.l4_cksum_pass) &&
1789 (!cqe->u0.s.ip_ver) &&
1790 (rq->lro.lro_cnt != 0)) {
1792 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1793 rq->lro_pkts_queued ++;
1796 /* If LRO posting fails then try to post to STACK */
1800 (*sc->ifp->if_input) (sc->ifp, m);
1801 #if defined(INET6) || defined(INET)
1804 /* Update rx stats per queue */
1805 rq->rx_stats.rx_pkts++;
1806 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1807 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1808 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1809 rq->rx_stats.rx_mcast_pkts++;
1810 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1811 rq->rx_stats.rx_ucast_pkts++;
1819 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1822 struct oce_packet_desc *pd;
1823 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1825 for (i = 0; i < num_frags; i++) {
1826 if (rq->ring->cidx == rq->ring->pidx) {
1827 device_printf(sc->dev,
1828 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1831 pd = &rq->pckts[rq->ring->cidx];
1832 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1833 bus_dmamap_unload(rq->tag, pd->map);
1834 if (pd->mbuf != NULL) {
1839 RING_GET(rq->ring, 1);
1846 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1848 struct oce_nic_rx_cqe_v1 *cqe_v1;
1851 if (sc->be3_native) {
1852 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1853 vtp = cqe_v1->u0.s.vlan_tag_present;
1855 vtp = cqe->u0.s.vlan_tag_present;
1863 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1865 struct oce_nic_rx_cqe_v1 *cqe_v1;
1868 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1869 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1870 port_id = cqe_v1->u0.s.port;
1871 if (sc->port_id != port_id)
1874 ;/* For BE3 legacy and Lancer this is dummy */
1880 #if defined(INET6) || defined(INET)
1882 oce_rx_flush_lro(struct oce_rq *rq)
1884 struct lro_ctrl *lro = &rq->lro;
1885 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1887 if (!IF_LRO_ENABLED(sc))
1890 tcp_lro_flush_all(lro);
1891 rq->lro_pkts_queued = 0;
1898 oce_init_lro(POCE_SOFTC sc)
1900 struct lro_ctrl *lro = NULL;
1903 for (i = 0; i < sc->nrqs; i++) {
1904 lro = &sc->rq[i]->lro;
1905 rc = tcp_lro_init(lro);
1907 device_printf(sc->dev, "LRO init failed\n");
1918 oce_free_lro(POCE_SOFTC sc)
1920 struct lro_ctrl *lro = NULL;
1923 for (i = 0; i < sc->nrqs; i++) {
1924 lro = &sc->rq[i]->lro;
1932 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1934 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1936 struct oce_packet_desc *pd;
1937 bus_dma_segment_t segs[6];
1938 int nsegs, added = 0;
1939 struct oce_nic_rqe *rqe;
1940 pd_rxulp_db_t rxdb_reg;
1942 uint32_t oce_max_rq_posts = 64;
1944 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1945 for (i = 0; i < count; i++) {
1946 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1948 pd = &rq->pckts[rq->ring->pidx];
1949 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1950 if (pd->mbuf == NULL) {
1951 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1954 pd->mbuf->m_nextpkt = NULL;
1956 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1958 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1961 segs, &nsegs, BUS_DMA_NOWAIT);
1964 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1973 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1975 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1976 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1977 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1978 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1979 RING_PUT(rq->ring, 1);
1983 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1985 for (i = added / oce_max_rq_posts; i > 0; i--) {
1986 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1987 rxdb_reg.bits.qid = rq->rq_id;
1989 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1990 val |= oce_max_rq_posts << 16;
1991 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1993 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1995 added -= oce_max_rq_posts;
1998 rxdb_reg.bits.qid = rq->rq_id;
1999 rxdb_reg.bits.num_posted = added;
2001 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
2003 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
2005 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
2014 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
2017 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
2018 if(!sc->enable_hwlro) {
2019 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
2020 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
2022 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
2023 oce_alloc_rx_bufs(rq, 64);
2031 oce_rq_handler_lro(void *arg)
2033 struct oce_rq *rq = (struct oce_rq *)arg;
2034 struct oce_cq *cq = rq->cq;
2035 POCE_SOFTC sc = rq->parent;
2036 struct nic_hwlro_singleton_cqe *cqe;
2037 struct nic_hwlro_cqe_part2 *cqe2;
2041 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2042 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2043 while (cqe->valid) {
2044 if(cqe->cqe_type == 0) { /* singleton cqe */
2045 /* we should not get singleton cqe after cqe1 on same rq */
2046 if(rq->cqe_firstpart != NULL) {
2047 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2048 goto exit_rq_handler_lro;
2050 if(cqe->error != 0) {
2051 rq->rx_stats.rxcp_err++;
2052 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2054 oce_rx_lro(rq, cqe, NULL);
2055 rq->rx_stats.rx_compl++;
2057 RING_GET(cq->ring, 1);
2059 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2061 }else if(cqe->cqe_type == 0x1) { /* first part */
2062 /* we should not get cqe1 after cqe1 on same rq */
2063 if(rq->cqe_firstpart != NULL) {
2064 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2065 goto exit_rq_handler_lro;
2067 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2068 RING_GET(cq->ring, 1);
2069 }else if(cqe->cqe_type == 0x2) { /* second part */
2070 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2071 if(cqe2->error != 0) {
2072 rq->rx_stats.rxcp_err++;
2073 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2075 /* We should not get cqe2 without cqe1 */
2076 if(rq->cqe_firstpart == NULL) {
2077 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2078 goto exit_rq_handler_lro;
2080 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2082 rq->rx_stats.rx_compl++;
2083 rq->cqe_firstpart->valid = 0;
2085 rq->cqe_firstpart = NULL;
2087 RING_GET(cq->ring, 1);
2089 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2093 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2094 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2096 oce_check_rx_bufs(sc, num_cqes, rq);
2097 exit_rq_handler_lro:
2098 UNLOCK(&rq->rx_lock);
2102 /* Handle the Completion Queue for receive */
2104 oce_rq_handler(void *arg)
2106 struct oce_rq *rq = (struct oce_rq *)arg;
2107 struct oce_cq *cq = rq->cq;
2108 POCE_SOFTC sc = rq->parent;
2109 struct oce_nic_rx_cqe *cqe;
2113 oce_rq_handler_lro(arg);
2117 bus_dmamap_sync(cq->ring->dma.tag,
2118 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2119 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2120 while (cqe->u0.dw[2]) {
2121 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2123 if (cqe->u0.s.error == 0) {
2126 rq->rx_stats.rxcp_err++;
2127 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2128 /* Post L3/L4 errors to stack.*/
2131 rq->rx_stats.rx_compl++;
2134 #if defined(INET6) || defined(INET)
2135 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2136 oce_rx_flush_lro(rq);
2140 RING_GET(cq->ring, 1);
2141 bus_dmamap_sync(cq->ring->dma.tag,
2142 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2144 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2146 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2150 #if defined(INET6) || defined(INET)
2151 if (IF_LRO_ENABLED(sc))
2152 oce_rx_flush_lro(rq);
2155 oce_check_rx_bufs(sc, num_cqes, rq);
2156 UNLOCK(&rq->rx_lock);
2164 /*****************************************************************************
2165 * Helper function prototypes in this file *
2166 *****************************************************************************/
2169 oce_attach_ifp(POCE_SOFTC sc)
2172 sc->ifp = if_alloc(IFT_ETHER);
2176 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2177 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2178 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2180 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2181 sc->ifp->if_ioctl = oce_ioctl;
2182 sc->ifp->if_start = oce_start;
2183 sc->ifp->if_init = oce_init;
2184 sc->ifp->if_mtu = ETHERMTU;
2185 sc->ifp->if_softc = sc;
2186 #if __FreeBSD_version >= 800000
2187 sc->ifp->if_transmit = oce_multiq_start;
2188 sc->ifp->if_qflush = oce_multiq_flush;
2191 if_initname(sc->ifp,
2192 device_get_name(sc->dev), device_get_unit(sc->dev));
2194 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2195 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2196 IFQ_SET_READY(&sc->ifp->if_snd);
2198 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2199 sc->ifp->if_hwassist |= CSUM_TSO;
2200 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2202 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2203 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2204 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2206 #if defined(INET6) || defined(INET)
2207 sc->ifp->if_capabilities |= IFCAP_TSO;
2208 sc->ifp->if_capabilities |= IFCAP_LRO;
2209 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2212 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2213 sc->ifp->if_baudrate = IF_Gbps(10);
2215 #if __FreeBSD_version >= 1000000
2216 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2217 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2218 sc->ifp->if_hw_tsomaxsegsize = 4096;
2221 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2228 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2230 POCE_SOFTC sc = ifp->if_softc;
2232 if (ifp->if_softc != arg)
2234 if ((vtag == 0) || (vtag > 4095))
2237 sc->vlan_tag[vtag] = 1;
2239 if (sc->vlans_added <= (sc->max_vlans + 1))
2245 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2247 POCE_SOFTC sc = ifp->if_softc;
2249 if (ifp->if_softc != arg)
2251 if ((vtag == 0) || (vtag > 4095))
2254 sc->vlan_tag[vtag] = 0;
2261 * A max of 64 vlans can be configured in BE. If the user configures
2262 * more, place the card in vlan promiscuous mode.
2265 oce_vid_config(POCE_SOFTC sc)
2267 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2268 uint16_t ntags = 0, i;
2271 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2272 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2273 for (i = 0; i < MAX_VLANS; i++) {
2274 if (sc->vlan_tag[i]) {
2275 vtags[ntags].vtag = i;
2280 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2281 vtags, ntags, 1, 0);
2283 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2290 oce_mac_addr_set(POCE_SOFTC sc)
2292 uint32_t old_pmac_id = sc->pmac_id;
2296 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2297 sc->macaddr.size_of_struct);
2301 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2302 sc->if_id, &sc->pmac_id);
2304 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2305 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2306 sc->macaddr.size_of_struct);
2309 device_printf(sc->dev, "Failed update macaddress\n");
2315 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2317 POCE_SOFTC sc = ifp->if_softc;
2318 struct ifreq *ifr = (struct ifreq *)data;
2320 char cookie[32] = {0};
2321 void *priv_data = ifr_data_get_ptr(ifr);
2325 OCE_DMA_MEM dma_mem;
2326 struct mbx_common_get_cntl_attr *fw_cmd;
2328 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2331 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2334 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2335 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2338 req_size = le32toh(req.u0.req.request_length);
2339 if (req_size > 65536)
2342 req_size += sizeof(struct mbx_hdr);
2343 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2347 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2352 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2358 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2362 firmware is filling all the attributes for this ioctl except
2363 the driver version..so fill it
2365 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2366 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2367 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2368 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2372 oce_dma_free(sc, &dma_mem);
2378 oce_eqd_set_periodic(POCE_SOFTC sc)
2380 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2381 struct oce_aic_obj *aic;
2383 uint64_t now = 0, delta;
2384 int eqd, i, num = 0;
2385 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2389 #define ticks_to_msecs(t) (1000 * (t) / hz)
2391 for (i = 0 ; i < sc->neqs; i++) {
2393 aic = &sc->aic_obj[i];
2394 /* When setting the static eq delay from the user space */
2404 rxpkts = rq->rx_stats.rx_pkts;
2407 if (i + 1 < sc->nrqs) {
2409 rxpkts += rq->rx_stats.rx_pkts;
2413 tx_reqs = wq->tx_stats.tx_reqs;
2418 if (!aic->ticks || now < aic->ticks ||
2419 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2420 aic->prev_rxpkts = rxpkts;
2421 aic->prev_txreqs = tx_reqs;
2426 delta = ticks_to_msecs(now - aic->ticks);
2428 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2429 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2430 eqd = (pps / 15000) << 2;
2434 /* Make sure that the eq delay is in the known range */
2435 eqd = min(eqd, aic->max_eqd);
2436 eqd = max(eqd, aic->min_eqd);
2438 aic->prev_rxpkts = rxpkts;
2439 aic->prev_txreqs = tx_reqs;
2443 if (eqd != aic->cur_eqd) {
2444 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2445 set_eqd[num].eq_id = eqo->eq_id;
2451 /* Is there atleast one eq that needs to be modified? */
2452 for(i = 0; i < num; i += 8) {
2454 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2456 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2461 static void oce_detect_hw_error(POCE_SOFTC sc)
2464 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2465 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2472 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2473 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2474 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2475 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2478 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2479 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2480 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2481 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2483 ue_low = (ue_low & ~ue_low_mask);
2484 ue_high = (ue_high & ~ue_high_mask);
2487 /* On certain platforms BE hardware can indicate spurious UEs.
2488 * Allow the h/w to stop working completely in case of a real UE.
2489 * Hence not setting the hw_error for UE detection.
2491 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2492 sc->hw_error = TRUE;
2493 device_printf(sc->dev, "Error detected in the card\n");
2496 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2497 device_printf(sc->dev,
2498 "ERR: sliport status 0x%x\n", sliport_status);
2499 device_printf(sc->dev,
2500 "ERR: sliport error1 0x%x\n", sliport_err1);
2501 device_printf(sc->dev,
2502 "ERR: sliport error2 0x%x\n", sliport_err2);
2506 for (i = 0; ue_low; ue_low >>= 1, i++) {
2508 device_printf(sc->dev, "UE: %s bit set\n",
2509 ue_status_low_desc[i]);
2514 for (i = 0; ue_high; ue_high >>= 1, i++) {
2516 device_printf(sc->dev, "UE: %s bit set\n",
2517 ue_status_hi_desc[i]);
2525 oce_local_timer(void *arg)
2527 POCE_SOFTC sc = arg;
2530 oce_detect_hw_error(sc);
2531 oce_refresh_nic_stats(sc);
2532 oce_refresh_queue_stats(sc);
2533 oce_mac_addr_set(sc);
2536 for (i = 0; i < sc->nwqs; i++)
2537 oce_tx_restart(sc, sc->wq[i]);
2539 /* calculate and set the eq delay for optimal interrupt rate */
2540 if (IS_BE(sc) || IS_SH(sc))
2541 oce_eqd_set_periodic(sc);
2543 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2547 oce_tx_compl_clean(POCE_SOFTC sc)
2550 int i = 0, timeo = 0, num_wqes = 0;
2551 int pending_txqs = sc->nwqs;
2553 /* Stop polling for compls when HW has been silent for 10ms or
2554 * hw_error or no outstanding completions expected
2557 pending_txqs = sc->nwqs;
2559 for_all_wq_queues(sc, wq, i) {
2560 num_wqes = oce_wq_handler(wq);
2565 if(!wq->ring->num_used)
2569 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2575 for_all_wq_queues(sc, wq, i) {
2576 while(wq->ring->num_used) {
2577 LOCK(&wq->tx_compl_lock);
2578 oce_process_tx_completion(wq);
2579 UNLOCK(&wq->tx_compl_lock);
2585 /* NOTE : This should only be called holding
2589 oce_if_deactivate(POCE_SOFTC sc)
2596 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2598 oce_tx_compl_clean(sc);
2600 /* Stop intrs and finish any bottom halves pending */
2601 oce_hw_intr_disable(sc);
2603 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2604 any other lock. So unlock device lock and require after
2605 completing taskqueue_drain.
2607 UNLOCK(&sc->dev_lock);
2608 for (i = 0; i < sc->intr_count; i++) {
2609 if (sc->intrs[i].tq != NULL) {
2610 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2613 LOCK(&sc->dev_lock);
2615 /* Delete RX queue in card with flush param */
2618 /* Invalidate any pending cq and eq entries*/
2619 for_all_evnt_queues(sc, eq, i)
2621 for_all_rq_queues(sc, rq, i)
2622 oce_drain_rq_cq(rq);
2623 for_all_wq_queues(sc, wq, i)
2624 oce_drain_wq_cq(wq);
2626 /* But still we need to get MCC aync events.
2627 So enable intrs and also arm first EQ
2629 oce_hw_intr_enable(sc);
2630 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2637 oce_if_activate(POCE_SOFTC sc)
2644 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2646 oce_hw_intr_disable(sc);
2650 for_all_rq_queues(sc, rq, i) {
2651 rc = oce_start_rq(rq);
2653 device_printf(sc->dev, "Unable to start RX\n");
2656 for_all_wq_queues(sc, wq, i) {
2657 rc = oce_start_wq(wq);
2659 device_printf(sc->dev, "Unable to start TX\n");
2663 for_all_evnt_queues(sc, eq, i)
2664 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2666 oce_hw_intr_enable(sc);
2671 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2673 /* Update Link status */
2674 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2675 ASYNC_EVENT_LINK_UP) {
2676 sc->link_status = ASYNC_EVENT_LINK_UP;
2677 if_link_state_change(sc->ifp, LINK_STATE_UP);
2679 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2680 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2685 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2686 struct oce_async_evt_grp5_os2bmc *evt)
2688 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2689 if (evt->u.s.mgmt_enable)
2690 sc->flags |= OCE_FLAGS_OS2BMC;
2694 sc->bmc_filt_mask = evt->u.s.arp_filter;
2695 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2696 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2697 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2698 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2699 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2700 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2701 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2702 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2706 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2708 struct oce_async_event_grp5_pvid_state *gcqe;
2709 struct oce_async_evt_grp5_os2bmc *bmccqe;
2711 switch (cqe->u0.s.async_type) {
2712 case ASYNC_EVENT_PVID_STATE:
2714 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2716 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2720 case ASYNC_EVENT_OS2BMC:
2721 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2722 oce_async_grp5_osbmc_process(sc, bmccqe);
2729 /* Handle the Completion Queue for the Mailbox/Async notifications */
2731 oce_mq_handler(void *arg)
2733 struct oce_mq *mq = (struct oce_mq *)arg;
2734 POCE_SOFTC sc = mq->parent;
2735 struct oce_cq *cq = mq->cq;
2736 int num_cqes = 0, evt_type = 0, optype = 0;
2737 struct oce_mq_cqe *cqe;
2738 struct oce_async_cqe_link_state *acqe;
2739 struct oce_async_event_qnq *dbgcqe;
2742 bus_dmamap_sync(cq->ring->dma.tag,
2743 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2744 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2746 while (cqe->u0.dw[3]) {
2747 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2748 if (cqe->u0.s.async_event) {
2749 evt_type = cqe->u0.s.event_type;
2750 optype = cqe->u0.s.async_type;
2751 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2752 /* Link status evt */
2753 acqe = (struct oce_async_cqe_link_state *)cqe;
2754 process_link_state(sc, acqe);
2755 } else if (evt_type == ASYNC_EVENT_GRP5) {
2756 oce_process_grp5_events(sc, cqe);
2757 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2758 optype == ASYNC_EVENT_DEBUG_QNQ) {
2759 dbgcqe = (struct oce_async_event_qnq *)cqe;
2761 sc->qnqid = dbgcqe->vlan_tag;
2762 sc->qnq_debug_event = TRUE;
2766 RING_GET(cq->ring, 1);
2767 bus_dmamap_sync(cq->ring->dma.tag,
2768 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2769 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2774 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2781 setup_max_queues_want(POCE_SOFTC sc)
2783 /* Check if it is FLEX machine. Is so dont use RSS */
2784 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2785 (sc->function_mode & FNM_UMC_MODE) ||
2786 (sc->function_mode & FNM_VNIC_MODE) ||
2787 (!is_rss_enabled(sc)) ||
2792 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2793 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2796 if (IS_BE2(sc) && is_rss_enabled(sc))
2797 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2802 update_queues_got(POCE_SOFTC sc)
2804 if (is_rss_enabled(sc)) {
2805 sc->nrqs = sc->intr_count + 1;
2806 sc->nwqs = sc->intr_count;
2817 oce_check_ipv6_ext_hdr(struct mbuf *m)
2819 struct ether_header *eh = mtod(m, struct ether_header *);
2820 caddr_t m_datatemp = m->m_data;
2822 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2823 m->m_data += sizeof(struct ether_header);
2824 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2826 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2827 (ip6->ip6_nxt != IPPROTO_UDP)){
2828 struct ip6_ext *ip6e = NULL;
2829 m->m_data += sizeof(struct ip6_hdr);
2831 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2832 if(ip6e->ip6e_len == 0xff) {
2833 m->m_data = m_datatemp;
2837 m->m_data = m_datatemp;
2843 is_be3_a1(POCE_SOFTC sc)
2845 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2851 static struct mbuf *
2852 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2854 uint16_t vlan_tag = 0;
2859 /* Embed vlan tag in the packet if it is not part of it */
2860 if(m->m_flags & M_VLANTAG) {
2861 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2862 m->m_flags &= ~M_VLANTAG;
2865 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2868 vlan_tag = sc->pvid;
2874 m = ether_vlanencap(m, vlan_tag);
2878 m = ether_vlanencap(m, sc->qnqid);
2887 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2889 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2890 oce_check_ipv6_ext_hdr(m)) {
2897 oce_get_config(POCE_SOFTC sc)
2900 uint32_t max_rss = 0;
2902 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2903 max_rss = OCE_LEGACY_MODE_RSS;
2905 max_rss = OCE_MAX_RSS;
2908 rc = oce_get_profile_config(sc, max_rss);
2910 sc->nwqs = OCE_MAX_WQ;
2911 sc->nrssqs = max_rss;
2912 sc->nrqs = sc->nrssqs + 1;
2915 else { /* For BE3 don't rely on fw for determining the resources */
2916 sc->nrssqs = max_rss;
2917 sc->nrqs = sc->nrssqs + 1;
2918 sc->nwqs = OCE_MAX_WQ;
2919 sc->max_vlans = MAX_VLANFILTER_SIZE;
2924 oce_rdma_close(void)
2926 if (oce_rdma_if != NULL) {
2932 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2934 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2938 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2941 struct oce_dev_info di;
2944 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2948 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2949 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2953 rdma_info->close = oce_rdma_close;
2954 rdma_info->mbox_post = oce_mbox_post;
2955 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2956 rdma_info->get_mac_addr = oce_get_mac_addr;
2958 oce_rdma_if = rdma_if;
2961 while (sc != NULL) {
2962 if (oce_rdma_if->announce != NULL) {
2963 memset(&di, 0, sizeof(di));
2967 di.db_bhandle = sc->db_bhandle;
2968 di.db_btag = sc->db_btag;
2969 di.db_page_size = 4096;
2970 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2971 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2972 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2973 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2975 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2977 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2978 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2979 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2980 di.msix.start_vector = sc->intr_count;
2981 for (i=0; i<di.msix.num_vectors; i++) {
2982 di.msix.vector_list[i] = sc->intrs[i].vector;
2986 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2987 di.vendor_id = pci_get_vendor(sc->dev);
2988 di.dev_id = pci_get_device(sc->dev);
2990 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2991 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2994 rdma_if->announce(&di);
3003 oce_read_env_variables( POCE_SOFTC sc )
3008 /* read if user wants to enable hwlro or swlro */
3009 //value = getenv("oce_enable_hwlro");
3010 if(value && IS_SH(sc)) {
3011 sc->enable_hwlro = strtol(value, NULL, 10);
3012 if(sc->enable_hwlro) {
3013 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
3015 device_printf(sc->dev, "no hardware lro support\n");
3016 device_printf(sc->dev, "software lro enabled\n");
3017 sc->enable_hwlro = 0;
3019 device_printf(sc->dev, "hardware lro enabled\n");
3020 oce_max_rsp_handled = 32;
3023 device_printf(sc->dev, "software lro enabled\n");
3026 sc->enable_hwlro = 0;
3029 /* read mbuf size */
3030 //value = getenv("oce_rq_buf_size");
3031 if(value && IS_SH(sc)) {
3032 oce_rq_buf_size = strtol(value, NULL, 10);
3033 switch(oce_rq_buf_size) {
3041 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
3042 oce_rq_buf_size = 2048;