2 * Copyright (C) 2013 Emulex
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are met:
8 * 1. Redistributions of source code must retain the above copyright notice,
9 * this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
15 * 3. Neither the name of the Emulex Corporation nor the names of its
16 * contributors may be used to endorse or promote products derived from
17 * this software without specific prior written permission.
19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
20 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
23 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29 * POSSIBILITY OF SUCH DAMAGE.
31 * Contact Information:
32 * freebsd-drivers@emulex.com
36 * Costa Mesa, CA 92626
41 #include "opt_inet6.h"
47 #define is_tso_pkt(m) (m->m_pkthdr.csum_flags & CSUM_TSO)
49 /* UE Status Low CSR */
50 static char *ue_status_low_desc[] = {
85 /* UE Status High CSR */
86 static char *ue_status_hi_desc[] = {
121 struct oce_common_cqe_info{
123 uint8_t l4_cksum_pass:1;
124 uint8_t ip_cksum_pass:1;
125 uint8_t ipv6_frame:1;
134 /* Driver entry points prototypes */
135 static int oce_probe(device_t dev);
136 static int oce_attach(device_t dev);
137 static int oce_detach(device_t dev);
138 static int oce_shutdown(device_t dev);
139 static int oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data);
140 static void oce_init(void *xsc);
141 static int oce_multiq_start(struct ifnet *ifp, struct mbuf *m);
142 static void oce_multiq_flush(struct ifnet *ifp);
144 /* Driver interrupt routines protypes */
145 static void oce_intr(void *arg, int pending);
146 static int oce_setup_intr(POCE_SOFTC sc);
147 static int oce_fast_isr(void *arg);
148 static int oce_alloc_intr(POCE_SOFTC sc, int vector,
149 void (*isr) (void *arg, int pending));
151 /* Media callbacks prototypes */
152 static void oce_media_status(struct ifnet *ifp, struct ifmediareq *req);
153 static int oce_media_change(struct ifnet *ifp);
155 /* Transmit routines prototypes */
156 static int oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index);
157 static void oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq);
158 static void oce_process_tx_completion(struct oce_wq *wq);
159 static int oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m,
162 /* Receive routines prototypes */
163 static int oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
164 static int oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe);
165 static void oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe);
166 static void oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq);
167 static uint16_t oce_rq_handler_lro(void *arg);
168 static void oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2);
169 static void oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2);
170 static void oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m);
172 /* Helper function prototypes in this file */
173 static int oce_attach_ifp(POCE_SOFTC sc);
174 static void oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
175 static void oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag);
176 static int oce_vid_config(POCE_SOFTC sc);
177 static void oce_mac_addr_set(POCE_SOFTC sc);
178 static int oce_handle_passthrough(struct ifnet *ifp, caddr_t data);
179 static void oce_local_timer(void *arg);
180 static void oce_if_deactivate(POCE_SOFTC sc);
181 static void oce_if_activate(POCE_SOFTC sc);
182 static void setup_max_queues_want(POCE_SOFTC sc);
183 static void update_queues_got(POCE_SOFTC sc);
184 static void process_link_state(POCE_SOFTC sc,
185 struct oce_async_cqe_link_state *acqe);
186 static int oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m);
187 static void oce_get_config(POCE_SOFTC sc);
188 static struct mbuf *oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete);
189 static void oce_read_env_variables(POCE_SOFTC sc);
193 #if defined(INET6) || defined(INET)
194 static int oce_init_lro(POCE_SOFTC sc);
195 static struct mbuf * oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp);
198 static device_method_t oce_dispatch[] = {
199 DEVMETHOD(device_probe, oce_probe),
200 DEVMETHOD(device_attach, oce_attach),
201 DEVMETHOD(device_detach, oce_detach),
202 DEVMETHOD(device_shutdown, oce_shutdown),
207 static driver_t oce_driver = {
212 static devclass_t oce_devclass;
215 DRIVER_MODULE(oce, pci, oce_driver, oce_devclass, 0, 0);
216 MODULE_DEPEND(oce, pci, 1, 1, 1);
217 MODULE_DEPEND(oce, ether, 1, 1, 1);
218 MODULE_VERSION(oce, 1);
222 const char component_revision[32] = {"///" COMPONENT_REVISION "///"};
224 /* Module capabilites and parameters */
225 uint32_t oce_max_rsp_handled = OCE_MAX_RSP_HANDLED;
226 uint32_t oce_enable_rss = OCE_MODCAP_RSS;
227 uint32_t oce_rq_buf_size = 2048;
229 TUNABLE_INT("hw.oce.max_rsp_handled", &oce_max_rsp_handled);
230 TUNABLE_INT("hw.oce.enable_rss", &oce_enable_rss);
233 /* Supported devices table */
234 static uint32_t supportedDevices[] = {
235 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE2,
236 (PCI_VENDOR_SERVERENGINES << 16) | PCI_PRODUCT_BE3,
237 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_BE3,
238 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201,
239 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_XE201_VF,
240 (PCI_VENDOR_EMULEX << 16) | PCI_PRODUCT_SH
243 POCE_SOFTC softc_head = NULL;
244 POCE_SOFTC softc_tail = NULL;
246 struct oce_rdma_if *oce_rdma_if = NULL;
248 /*****************************************************************************
249 * Driver entry points functions *
250 *****************************************************************************/
253 oce_probe(device_t dev)
261 sc = device_get_softc(dev);
262 bzero(sc, sizeof(OCE_SOFTC));
265 vendor = pci_get_vendor(dev);
266 device = pci_get_device(dev);
268 for (i = 0; i < (sizeof(supportedDevices) / sizeof(uint32_t)); i++) {
269 if (vendor == ((supportedDevices[i] >> 16) & 0xffff)) {
270 if (device == (supportedDevices[i] & 0xffff)) {
271 sprintf(str, "%s:%s", "Emulex CNA NIC function",
273 device_set_desc_copy(dev, str);
276 case PCI_PRODUCT_BE2:
277 sc->flags |= OCE_FLAGS_BE2;
279 case PCI_PRODUCT_BE3:
280 sc->flags |= OCE_FLAGS_BE3;
282 case PCI_PRODUCT_XE201:
283 case PCI_PRODUCT_XE201_VF:
284 sc->flags |= OCE_FLAGS_XE201;
287 sc->flags |= OCE_FLAGS_SH;
292 return BUS_PROBE_DEFAULT;
302 oce_attach(device_t dev)
307 sc = device_get_softc(dev);
309 rc = oce_hw_pci_alloc(sc);
313 sc->tx_ring_size = OCE_TX_RING_SIZE;
314 sc->rx_ring_size = OCE_RX_RING_SIZE;
315 /* receive fragment size should be multiple of 2K */
316 sc->rq_frag_size = ((oce_rq_buf_size / 2048) * 2048);
317 sc->flow_control = OCE_DEFAULT_FLOW_CONTROL;
318 sc->promisc = OCE_DEFAULT_PROMISCUOUS;
320 LOCK_CREATE(&sc->bmbx_lock, "Mailbox_lock");
321 LOCK_CREATE(&sc->dev_lock, "Device_lock");
323 /* initialise the hardware */
324 rc = oce_hw_init(sc);
328 oce_read_env_variables(sc);
332 setup_max_queues_want(sc);
334 rc = oce_setup_intr(sc);
338 rc = oce_queue_init_all(sc);
342 rc = oce_attach_ifp(sc);
346 #if defined(INET6) || defined(INET)
347 rc = oce_init_lro(sc);
352 rc = oce_hw_start(sc);
356 sc->vlan_attach = EVENTHANDLER_REGISTER(vlan_config,
357 oce_add_vlan, sc, EVENTHANDLER_PRI_FIRST);
358 sc->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig,
359 oce_del_vlan, sc, EVENTHANDLER_PRI_FIRST);
361 rc = oce_stats_init(sc);
367 callout_init(&sc->timer, CALLOUT_MPSAFE);
368 rc = callout_reset(&sc->timer, 2 * hz, oce_local_timer, sc);
373 if (softc_tail != NULL) {
374 softc_tail->next = sc;
383 callout_drain(&sc->timer);
387 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
389 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
390 oce_hw_intr_disable(sc);
392 #if defined(INET6) || defined(INET)
396 ether_ifdetach(sc->ifp);
399 oce_queue_release_all(sc);
403 oce_dma_free(sc, &sc->bsmbx);
406 LOCK_DESTROY(&sc->dev_lock);
407 LOCK_DESTROY(&sc->bmbx_lock);
414 oce_detach(device_t dev)
416 POCE_SOFTC sc = device_get_softc(dev);
417 POCE_SOFTC poce_sc_tmp, *ppoce_sc_tmp1, poce_sc_tmp2 = NULL;
419 poce_sc_tmp = softc_head;
420 ppoce_sc_tmp1 = &softc_head;
421 while (poce_sc_tmp != NULL) {
422 if (poce_sc_tmp == sc) {
423 *ppoce_sc_tmp1 = sc->next;
424 if (sc->next == NULL) {
425 softc_tail = poce_sc_tmp2;
429 poce_sc_tmp2 = poce_sc_tmp;
430 ppoce_sc_tmp1 = &poce_sc_tmp->next;
431 poce_sc_tmp = poce_sc_tmp->next;
435 oce_if_deactivate(sc);
436 UNLOCK(&sc->dev_lock);
438 callout_drain(&sc->timer);
440 if (sc->vlan_attach != NULL)
441 EVENTHANDLER_DEREGISTER(vlan_config, sc->vlan_attach);
442 if (sc->vlan_detach != NULL)
443 EVENTHANDLER_DEREGISTER(vlan_unconfig, sc->vlan_detach);
445 ether_ifdetach(sc->ifp);
451 bus_generic_detach(dev);
458 oce_shutdown(device_t dev)
462 rc = oce_detach(dev);
469 oce_ioctl(struct ifnet *ifp, u_long command, caddr_t data)
471 struct ifreq *ifr = (struct ifreq *)data;
472 POCE_SOFTC sc = ifp->if_softc;
481 rc = ifmedia_ioctl(ifp, ifr, &sc->media, command);
485 if (ifr->ifr_mtu > OCE_MAX_MTU)
488 ifp->if_mtu = ifr->ifr_mtu;
492 if (ifp->if_flags & IFF_UP) {
493 if (!(ifp->if_drv_flags & IFF_DRV_RUNNING)) {
494 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
497 device_printf(sc->dev, "Interface Up\n");
501 sc->ifp->if_drv_flags &=
502 ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
503 oce_if_deactivate(sc);
505 UNLOCK(&sc->dev_lock);
507 device_printf(sc->dev, "Interface Down\n");
510 if ((ifp->if_flags & IFF_PROMISC) && !sc->promisc) {
511 if (!oce_rxf_set_promiscuous(sc, (1 | (1 << 1))))
513 } else if (!(ifp->if_flags & IFF_PROMISC) && sc->promisc) {
514 if (!oce_rxf_set_promiscuous(sc, 0))
522 rc = oce_hw_update_multicast(sc);
524 device_printf(sc->dev,
525 "Update multicast address failed\n");
529 u = ifr->ifr_reqcap ^ ifp->if_capenable;
531 if (u & IFCAP_TXCSUM) {
532 ifp->if_capenable ^= IFCAP_TXCSUM;
533 ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP);
535 if (IFCAP_TSO & ifp->if_capenable &&
536 !(IFCAP_TXCSUM & ifp->if_capenable)) {
537 ifp->if_capenable &= ~IFCAP_TSO;
538 ifp->if_hwassist &= ~CSUM_TSO;
540 "TSO disabled due to -txcsum.\n");
544 if (u & IFCAP_RXCSUM)
545 ifp->if_capenable ^= IFCAP_RXCSUM;
547 if (u & IFCAP_TSO4) {
548 ifp->if_capenable ^= IFCAP_TSO4;
550 if (IFCAP_TSO & ifp->if_capenable) {
551 if (IFCAP_TXCSUM & ifp->if_capenable)
552 ifp->if_hwassist |= CSUM_TSO;
554 ifp->if_capenable &= ~IFCAP_TSO;
555 ifp->if_hwassist &= ~CSUM_TSO;
557 "Enable txcsum first.\n");
561 ifp->if_hwassist &= ~CSUM_TSO;
564 if (u & IFCAP_VLAN_HWTAGGING)
565 ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING;
567 if (u & IFCAP_VLAN_HWFILTER) {
568 ifp->if_capenable ^= IFCAP_VLAN_HWFILTER;
571 #if defined(INET6) || defined(INET)
573 ifp->if_capenable ^= IFCAP_LRO;
574 if(sc->enable_hwlro) {
575 if(ifp->if_capenable & IFCAP_LRO) {
576 rc = oce_mbox_nic_set_iface_lro_config(sc, 1);
578 rc = oce_mbox_nic_set_iface_lro_config(sc, 0);
587 rc = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c));
591 if (i2c.dev_addr != PAGE_NUM_A0 &&
592 i2c.dev_addr != PAGE_NUM_A2) {
597 if (i2c.len > sizeof(i2c.data)) {
602 rc = oce_mbox_read_transrecv_data(sc, i2c.dev_addr);
608 if (i2c.dev_addr == PAGE_NUM_A0)
611 offset = TRANSCEIVER_A0_SIZE + i2c.offset;
613 memcpy(&i2c.data[0], &sfp_vpd_dump_buffer[offset], i2c.len);
615 rc = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c));
619 rc = priv_check(curthread, PRIV_DRIVER);
622 rc = oce_handle_passthrough(ifp, data);
625 rc = ether_ioctl(ifp, command, data);
640 if (sc->ifp->if_flags & IFF_UP) {
641 oce_if_deactivate(sc);
645 UNLOCK(&sc->dev_lock);
651 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
653 POCE_SOFTC sc = ifp->if_softc;
654 struct oce_wq *wq = NULL;
658 if (!sc->link_status)
661 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
662 queue_index = m->m_pkthdr.flowid % sc->nwqs;
664 wq = sc->wq[queue_index];
667 status = oce_multiq_transmit(ifp, m, wq);
668 UNLOCK(&wq->tx_lock);
676 oce_multiq_flush(struct ifnet *ifp)
678 POCE_SOFTC sc = ifp->if_softc;
682 for (i = 0; i < sc->nwqs; i++) {
683 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
691 /*****************************************************************************
692 * Driver interrupt routines functions *
693 *****************************************************************************/
696 oce_intr(void *arg, int pending)
699 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
700 POCE_SOFTC sc = ii->sc;
701 struct oce_eq *eq = ii->eq;
703 struct oce_cq *cq = NULL;
707 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
708 BUS_DMASYNC_POSTWRITE);
710 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
714 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
715 BUS_DMASYNC_POSTWRITE);
716 RING_GET(eq->ring, 1);
722 goto eq_arm; /* Spurious */
724 /* Clear EQ entries, but dont arm */
725 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
727 /* Process TX, RX and MCC. But dont arm CQ*/
728 for (i = 0; i < eq->cq_valid; i++) {
730 (*cq->cq_handler)(cq->cb_arg);
733 /* Arm all cqs connected to this EQ */
734 for (i = 0; i < eq->cq_valid; i++) {
736 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
740 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
747 oce_setup_intr(POCE_SOFTC sc)
749 int rc = 0, use_intx = 0;
750 int vector = 0, req_vectors = 0;
751 int tot_req_vectors, tot_vectors;
753 if (is_rss_enabled(sc))
754 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
758 tot_req_vectors = req_vectors;
759 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
760 if (req_vectors > 1) {
761 tot_req_vectors += OCE_RDMA_VECTORS;
762 sc->roce_intr_count = OCE_RDMA_VECTORS;
766 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
767 sc->intr_count = req_vectors;
768 tot_vectors = tot_req_vectors;
769 rc = pci_alloc_msix(sc->dev, &tot_vectors);
772 pci_release_msi(sc->dev);
774 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
775 if (tot_vectors < tot_req_vectors) {
776 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
777 sc->roce_intr_count = (tot_vectors / 2);
779 sc->intr_count = tot_vectors - sc->roce_intr_count;
782 sc->intr_count = tot_vectors;
784 sc->flags |= OCE_FLAGS_USING_MSIX;
792 /* Scale number of queues based on intr we got */
793 update_queues_got(sc);
796 device_printf(sc->dev, "Using legacy interrupt\n");
797 rc = oce_alloc_intr(sc, vector, oce_intr);
801 for (; vector < sc->intr_count; vector++) {
802 rc = oce_alloc_intr(sc, vector, oce_intr);
816 oce_fast_isr(void *arg)
818 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
819 POCE_SOFTC sc = ii->sc;
824 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
826 taskqueue_enqueue(ii->tq, &ii->task);
830 return FILTER_HANDLED;
835 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
837 POCE_INTR_INFO ii = &sc->intrs[vector];
840 if (vector >= OCE_MAX_EQ)
843 /* Set the resource id for the interrupt.
844 * MSIx is vector + 1 for the resource id,
845 * INTx is 0 for the resource id.
847 if (sc->flags & OCE_FLAGS_USING_MSIX)
851 ii->intr_res = bus_alloc_resource_any(sc->dev,
853 &rr, RF_ACTIVE|RF_SHAREABLE);
855 if (ii->intr_res == NULL) {
856 device_printf(sc->dev,
857 "Could not allocate interrupt\n");
862 TASK_INIT(&ii->task, 0, isr, ii);
864 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
865 ii->tq = taskqueue_create_fast(ii->task_name,
867 taskqueue_thread_enqueue,
869 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
870 device_get_nameunit(sc->dev));
873 rc = bus_setup_intr(sc->dev,
876 oce_fast_isr, NULL, ii, &ii->tag);
883 oce_intr_free(POCE_SOFTC sc)
887 for (i = 0; i < sc->intr_count; i++) {
889 if (sc->intrs[i].tag != NULL)
890 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
892 if (sc->intrs[i].tq != NULL)
893 taskqueue_free(sc->intrs[i].tq);
895 if (sc->intrs[i].intr_res != NULL)
896 bus_release_resource(sc->dev, SYS_RES_IRQ,
898 sc->intrs[i].intr_res);
899 sc->intrs[i].tag = NULL;
900 sc->intrs[i].intr_res = NULL;
903 if (sc->flags & OCE_FLAGS_USING_MSIX)
904 pci_release_msi(sc->dev);
910 /******************************************************************************
911 * Media callbacks functions *
912 ******************************************************************************/
915 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
917 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
920 req->ifm_status = IFM_AVALID;
921 req->ifm_active = IFM_ETHER;
923 if (sc->link_status == 1)
924 req->ifm_status |= IFM_ACTIVE;
928 switch (sc->link_speed) {
929 case 1: /* 10 Mbps */
930 req->ifm_active |= IFM_10_T | IFM_FDX;
933 case 2: /* 100 Mbps */
934 req->ifm_active |= IFM_100_TX | IFM_FDX;
938 req->ifm_active |= IFM_1000_T | IFM_FDX;
941 case 4: /* 10 Gbps */
942 req->ifm_active |= IFM_10G_SR | IFM_FDX;
945 case 5: /* 20 Gbps */
946 req->ifm_active |= IFM_10G_SR | IFM_FDX;
949 case 6: /* 25 Gbps */
950 req->ifm_active |= IFM_10G_SR | IFM_FDX;
953 case 7: /* 40 Gbps */
954 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
967 oce_media_change(struct ifnet *ifp)
973 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
974 struct mbuf *m, boolean_t *os2bmc,
977 struct ether_header *eh = NULL;
979 eh = mtod(m, struct ether_header *);
981 if (!is_os2bmc_enabled(sc) || *os2bmc) {
985 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
988 if (is_mc_allowed_on_bmc(sc, eh) ||
989 is_bc_allowed_on_bmc(sc, eh) ||
990 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
995 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
996 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
997 uint8_t nexthdr = ip6->ip6_nxt;
998 if (nexthdr == IPPROTO_ICMPV6) {
999 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
1000 switch (icmp6->icmp6_type) {
1001 case ND_ROUTER_ADVERT:
1002 *os2bmc = is_ipv6_ra_filt_enabled(sc);
1004 case ND_NEIGHBOR_ADVERT:
1005 *os2bmc = is_ipv6_na_filt_enabled(sc);
1013 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
1014 struct ip *ip = mtod(m, struct ip *);
1015 int iphlen = ip->ip_hl << 2;
1016 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
1017 switch (uh->uh_dport) {
1018 case DHCP_CLIENT_PORT:
1019 *os2bmc = is_dhcp_client_filt_enabled(sc);
1021 case DHCP_SERVER_PORT:
1022 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
1024 case NET_BIOS_PORT1:
1025 case NET_BIOS_PORT2:
1026 *os2bmc = is_nbios_filt_enabled(sc);
1028 case DHCPV6_RAS_PORT:
1029 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1037 *m_new = m_dup(m, M_NOWAIT);
1042 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1048 /*****************************************************************************
1049 * Transmit routines functions *
1050 *****************************************************************************/
1053 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1055 int rc = 0, i, retry_cnt = 0;
1056 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1057 struct mbuf *m, *m_temp, *m_new = NULL;
1058 struct oce_wq *wq = sc->wq[wq_index];
1059 struct oce_packet_desc *pd;
1060 struct oce_nic_hdr_wqe *nichdr;
1061 struct oce_nic_frag_wqe *nicfrag;
1062 struct ether_header *eh = NULL;
1065 boolean_t complete = TRUE;
1066 boolean_t os2bmc = FALSE;
1072 if (!(m->m_flags & M_PKTHDR)) {
1077 /* Don't allow non-TSO packets longer than MTU */
1078 if (!is_tso_pkt(m)) {
1079 eh = mtod(m, struct ether_header *);
1080 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1084 if(oce_tx_asic_stall_verify(sc, m)) {
1085 m = oce_insert_vlan_tag(sc, m, &complete);
1087 device_printf(sc->dev, "Insertion unsuccessful\n");
1093 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1094 * may cause a transmit stall on that port. So the work-around is to
1095 * pad short packets (<= 32 bytes) to a 36-byte length.
1097 if(IS_SH(sc) || IS_XE201(sc) ) {
1098 if(m->m_pkthdr.len <= 32) {
1100 bzero((void *)buf, 36);
1101 m_append(m, (36 - m->m_pkthdr.len), buf);
1106 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1107 /* consolidate packet buffers for TSO/LSO segment offload */
1108 #if defined(INET6) || defined(INET)
1109 m = oce_tso_setup(sc, mpp);
1120 pd = &wq->pckts[wq->pkt_desc_head];
1123 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1125 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1127 num_wqes = pd->nsegs + 1;
1128 if (IS_BE(sc) || IS_SH(sc)) {
1129 /*Dummy required only for BE3.*/
1133 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1134 bus_dmamap_unload(wq->tag, pd->map);
1137 atomic_store_rel_int(&wq->pkt_desc_head,
1138 (wq->pkt_desc_head + 1) % \
1139 OCE_WQ_PACKET_ARRAY_SIZE);
1140 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1144 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1145 nichdr->u0.dw[0] = 0;
1146 nichdr->u0.dw[1] = 0;
1147 nichdr->u0.dw[2] = 0;
1148 nichdr->u0.dw[3] = 0;
1150 nichdr->u0.s.complete = complete;
1151 nichdr->u0.s.mgmt = os2bmc;
1152 nichdr->u0.s.event = 1;
1153 nichdr->u0.s.crc = 1;
1154 nichdr->u0.s.forward = 0;
1155 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1156 nichdr->u0.s.udpcs =
1157 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1158 nichdr->u0.s.tcpcs =
1159 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1160 nichdr->u0.s.num_wqe = num_wqes;
1161 nichdr->u0.s.total_length = m->m_pkthdr.len;
1163 if (m->m_flags & M_VLANTAG) {
1164 nichdr->u0.s.vlan = 1; /*Vlan present*/
1165 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1168 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1169 if (m->m_pkthdr.tso_segsz) {
1170 nichdr->u0.s.lso = 1;
1171 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1173 if (!IS_BE(sc) || !IS_SH(sc))
1174 nichdr->u0.s.ipcs = 1;
1177 RING_PUT(wq->ring, 1);
1178 atomic_add_int(&wq->ring->num_used, 1);
1180 for (i = 0; i < pd->nsegs; i++) {
1182 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1183 struct oce_nic_frag_wqe);
1184 nicfrag->u0.s.rsvd0 = 0;
1185 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1186 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1187 nicfrag->u0.s.frag_len = segs[i].ds_len;
1188 pd->wqe_idx = wq->ring->pidx;
1189 RING_PUT(wq->ring, 1);
1190 atomic_add_int(&wq->ring->num_used, 1);
1192 if (num_wqes > (pd->nsegs + 1)) {
1194 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1195 struct oce_nic_frag_wqe);
1196 nicfrag->u0.dw[0] = 0;
1197 nicfrag->u0.dw[1] = 0;
1198 nicfrag->u0.dw[2] = 0;
1199 nicfrag->u0.dw[3] = 0;
1200 pd->wqe_idx = wq->ring->pidx;
1201 RING_PUT(wq->ring, 1);
1202 atomic_add_int(&wq->ring->num_used, 1);
1206 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1207 wq->tx_stats.tx_reqs++;
1208 wq->tx_stats.tx_wrbs += num_wqes;
1209 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1210 wq->tx_stats.tx_pkts++;
1212 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1213 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1214 reg_value = (num_wqes << 16) | wq->wq_id;
1216 /* if os2bmc is not enabled or if the pkt is already tagged as
1219 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1221 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1223 } else if (rc == EFBIG) {
1224 if (retry_cnt == 0) {
1225 m_temp = m_defrag(m, M_NOWAIT);
1230 retry_cnt = retry_cnt + 1;
1234 } else if (rc == ENOMEM)
1254 oce_process_tx_completion(struct oce_wq *wq)
1256 struct oce_packet_desc *pd;
1257 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1260 pd = &wq->pckts[wq->pkt_desc_tail];
1261 atomic_store_rel_int(&wq->pkt_desc_tail,
1262 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1263 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1264 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1265 bus_dmamap_unload(wq->tag, pd->map);
1272 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1273 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1274 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1275 oce_tx_restart(sc, wq);
1282 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1285 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1288 #if __FreeBSD_version >= 800000
1289 if (!drbr_empty(sc->ifp, wq->br))
1291 if (!IFQ_DRV_IS_EMPTY(&sc->ifp->if_snd))
1293 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1298 #if defined(INET6) || defined(INET)
1299 static struct mbuf *
1300 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1307 struct ip6_hdr *ip6;
1309 struct ether_vlan_header *eh;
1312 int total_len = 0, ehdrlen = 0;
1316 if (M_WRITABLE(m) == 0) {
1317 m = m_dup(*mpp, M_NOWAIT);
1324 eh = mtod(m, struct ether_vlan_header *);
1325 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1326 etype = ntohs(eh->evl_proto);
1327 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1329 etype = ntohs(eh->evl_encap_proto);
1330 ehdrlen = ETHER_HDR_LEN;
1336 ip = (struct ip *)(m->m_data + ehdrlen);
1337 if (ip->ip_p != IPPROTO_TCP)
1339 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1341 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1345 case ETHERTYPE_IPV6:
1346 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1347 if (ip6->ip6_nxt != IPPROTO_TCP)
1349 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1351 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1358 m = m_pullup(m, total_len);
1365 #endif /* INET6 || INET */
1368 oce_tx_task(void *arg, int npending)
1370 struct oce_wq *wq = arg;
1371 POCE_SOFTC sc = wq->parent;
1372 struct ifnet *ifp = sc->ifp;
1375 #if __FreeBSD_version >= 800000
1377 rc = oce_multiq_transmit(ifp, NULL, wq);
1379 device_printf(sc->dev,
1380 "TX[%d] restart failed\n", wq->queue_index);
1382 UNLOCK(&wq->tx_lock);
1391 oce_start(struct ifnet *ifp)
1393 POCE_SOFTC sc = ifp->if_softc;
1396 int def_q = 0; /* Defualt tx queue is 0*/
1398 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1402 if (!sc->link_status)
1406 IF_DEQUEUE(&sc->ifp->if_snd, m);
1410 LOCK(&sc->wq[def_q]->tx_lock);
1411 rc = oce_tx(sc, &m, def_q);
1412 UNLOCK(&sc->wq[def_q]->tx_lock);
1415 sc->wq[def_q]->tx_stats.tx_stops ++;
1416 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1417 IFQ_DRV_PREPEND(&ifp->if_snd, m);
1423 ETHER_BPF_MTAP(ifp, m);
1431 /* Handle the Completion Queue for transmit */
1433 oce_wq_handler(void *arg)
1435 struct oce_wq *wq = (struct oce_wq *)arg;
1436 POCE_SOFTC sc = wq->parent;
1437 struct oce_cq *cq = wq->cq;
1438 struct oce_nic_tx_cqe *cqe;
1441 LOCK(&wq->tx_compl_lock);
1442 bus_dmamap_sync(cq->ring->dma.tag,
1443 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1444 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1445 while (cqe->u0.dw[3]) {
1446 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe));
1448 wq->ring->cidx = cqe->u0.s.wqe_index + 1;
1449 if (wq->ring->cidx >= wq->ring->num_items)
1450 wq->ring->cidx -= wq->ring->num_items;
1452 oce_process_tx_completion(wq);
1453 wq->tx_stats.tx_compl++;
1455 RING_GET(cq->ring, 1);
1456 bus_dmamap_sync(cq->ring->dma.tag,
1457 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1459 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1464 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
1466 UNLOCK(&wq->tx_compl_lock);
1472 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq)
1474 POCE_SOFTC sc = ifp->if_softc;
1475 int status = 0, queue_index = 0;
1476 struct mbuf *next = NULL;
1477 struct buf_ring *br = NULL;
1480 queue_index = wq->queue_index;
1482 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1485 status = drbr_enqueue(ifp, br, m);
1490 if ((status = drbr_enqueue(ifp, br, m)) != 0)
1493 while ((next = drbr_peek(ifp, br)) != NULL) {
1494 if (oce_tx(sc, &next, queue_index)) {
1496 drbr_advance(ifp, br);
1498 drbr_putback(ifp, br, next);
1499 wq->tx_stats.tx_stops ++;
1500 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1504 drbr_advance(ifp, br);
1505 if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
1506 if (next->m_flags & M_MCAST)
1507 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1508 ETHER_BPF_MTAP(ifp, next);
1517 /*****************************************************************************
1518 * Receive routines functions *
1519 *****************************************************************************/
1522 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2)
1525 struct ether_header *eh = NULL;
1526 struct tcphdr *tcp_hdr = NULL;
1527 struct ip *ip4_hdr = NULL;
1528 struct ip6_hdr *ip6 = NULL;
1529 uint32_t payload_len = 0;
1531 eh = mtod(m, struct ether_header *);
1532 /* correct IP header */
1533 if(!cqe2->ipv6_frame) {
1534 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header));
1535 ip4_hdr->ip_ttl = cqe2->frame_lifespan;
1536 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header));
1537 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip));
1539 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header));
1540 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan;
1541 payload_len = cqe2->coalesced_size - sizeof(struct ether_header)
1542 - sizeof(struct ip6_hdr);
1543 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len);
1544 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr));
1547 /* correct tcp header */
1548 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num);
1550 tcp_hdr->th_flags |= TH_PUSH;
1552 tcp_hdr->th_win = htons(cqe2->tcp_window);
1553 tcp_hdr->th_sum = 0xffff;
1555 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2);
1556 *p = cqe1->tcp_timestamp_val;
1557 *(p+1) = cqe1->tcp_timestamp_ecr;
1564 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m)
1566 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1567 uint32_t i = 0, frag_len = 0;
1568 uint32_t len = cqe_info->pkt_size;
1569 struct oce_packet_desc *pd;
1570 struct mbuf *tail = NULL;
1572 for (i = 0; i < cqe_info->num_frags; i++) {
1573 if (rq->ring->cidx == rq->ring->pidx) {
1574 device_printf(sc->dev,
1575 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n");
1578 pd = &rq->pckts[rq->ring->cidx];
1580 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1581 bus_dmamap_unload(rq->tag, pd->map);
1582 RING_GET(rq->ring, 1);
1585 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len;
1586 pd->mbuf->m_len = frag_len;
1589 /* additional fragments */
1590 pd->mbuf->m_flags &= ~M_PKTHDR;
1591 tail->m_next = pd->mbuf;
1593 tail->m_nextpkt = NULL;
1596 /* first fragment, fill out much of the packet header */
1597 pd->mbuf->m_pkthdr.len = len;
1599 pd->mbuf->m_nextpkt = NULL;
1600 pd->mbuf->m_pkthdr.csum_flags = 0;
1601 if (IF_CSUM_ENABLED(sc)) {
1602 if (cqe_info->l4_cksum_pass) {
1603 if(!cqe_info->ipv6_frame) { /* IPV4 */
1604 pd->mbuf->m_pkthdr.csum_flags |=
1605 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1606 }else { /* IPV6 frame */
1608 pd->mbuf->m_pkthdr.csum_flags |=
1609 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1612 pd->mbuf->m_pkthdr.csum_data = 0xffff;
1614 if (cqe_info->ip_cksum_pass) {
1615 pd->mbuf->m_pkthdr.csum_flags |=
1616 (CSUM_IP_CHECKED|CSUM_IP_VALID);
1619 *m = tail = pd->mbuf;
1629 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2)
1631 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1632 struct nic_hwlro_cqe_part1 *cqe1 = NULL;
1633 struct mbuf *m = NULL;
1634 struct oce_common_cqe_info cq_info;
1638 cq_info.pkt_size = cqe->pkt_size;
1639 cq_info.vtag = cqe->vlan_tag;
1640 cq_info.l4_cksum_pass = cqe->l4_cksum_pass;
1641 cq_info.ip_cksum_pass = cqe->ip_cksum_pass;
1642 cq_info.ipv6_frame = cqe->ipv6_frame;
1643 cq_info.vtp = cqe->vtp;
1644 cq_info.qnq = cqe->qnq;
1646 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe;
1647 cq_info.pkt_size = cqe2->coalesced_size;
1648 cq_info.vtag = cqe2->vlan_tag;
1649 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass;
1650 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass;
1651 cq_info.ipv6_frame = cqe2->ipv6_frame;
1652 cq_info.vtp = cqe2->vtp;
1653 cq_info.qnq = cqe1->qnq;
1656 cq_info.vtag = BSWAP_16(cq_info.vtag);
1658 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size;
1659 if(cq_info.pkt_size % rq->cfg.frag_size)
1660 cq_info.num_frags++;
1662 oce_rx_mbuf_chain(rq, &cq_info, &m);
1666 //assert(cqe2->valid != 0);
1668 //assert(cqe2->cqe_type != 2);
1669 oce_correct_header(m, cqe1, cqe2);
1672 m->m_pkthdr.rcvif = sc->ifp;
1673 #if __FreeBSD_version >= 800000
1674 if (rq->queue_index)
1675 m->m_pkthdr.flowid = (rq->queue_index - 1);
1677 m->m_pkthdr.flowid = rq->queue_index;
1678 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1680 /* This deternies if vlan tag is Valid */
1682 if (sc->function_mode & FNM_FLEX10_MODE) {
1683 /* FLEX10. If QnQ is not set, neglect VLAN */
1685 m->m_pkthdr.ether_vtag = cq_info.vtag;
1686 m->m_flags |= M_VLANTAG;
1688 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1689 /* In UMC mode generally pvid will be striped by
1690 hw. But in some cases we have seen it comes
1691 with pvid. So if pvid == vlan, neglect vlan.
1693 m->m_pkthdr.ether_vtag = cq_info.vtag;
1694 m->m_flags |= M_VLANTAG;
1697 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1699 (*sc->ifp->if_input) (sc->ifp, m);
1701 /* Update rx stats per queue */
1702 rq->rx_stats.rx_pkts++;
1703 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1704 rq->rx_stats.rx_frags += cq_info.num_frags;
1705 rq->rx_stats.rx_ucast_pkts++;
1711 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1713 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1715 struct mbuf *m = NULL;
1716 struct oce_common_cqe_info cq_info;
1719 /* Is it a flush compl that has no data */
1720 if(!cqe->u0.s.num_fragments)
1723 len = cqe->u0.s.pkt_size;
1725 /*partial DMA workaround for Lancer*/
1726 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1730 if (!oce_cqe_portid_valid(sc, cqe)) {
1731 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1735 /* Get vlan_tag value */
1736 if(IS_BE(sc) || IS_SH(sc))
1737 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1739 vtag = cqe->u0.s.vlan_tag;
1741 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1742 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1743 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1744 cq_info.num_frags = cqe->u0.s.num_fragments;
1745 cq_info.pkt_size = cqe->u0.s.pkt_size;
1747 oce_rx_mbuf_chain(rq, &cq_info, &m);
1750 m->m_pkthdr.rcvif = sc->ifp;
1751 #if __FreeBSD_version >= 800000
1752 if (rq->queue_index)
1753 m->m_pkthdr.flowid = (rq->queue_index - 1);
1755 m->m_pkthdr.flowid = rq->queue_index;
1756 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1758 /* This deternies if vlan tag is Valid */
1759 if (oce_cqe_vtp_valid(sc, cqe)) {
1760 if (sc->function_mode & FNM_FLEX10_MODE) {
1761 /* FLEX10. If QnQ is not set, neglect VLAN */
1762 if (cqe->u0.s.qnq) {
1763 m->m_pkthdr.ether_vtag = vtag;
1764 m->m_flags |= M_VLANTAG;
1766 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1767 /* In UMC mode generally pvid will be striped by
1768 hw. But in some cases we have seen it comes
1769 with pvid. So if pvid == vlan, neglect vlan.
1771 m->m_pkthdr.ether_vtag = vtag;
1772 m->m_flags |= M_VLANTAG;
1776 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1777 #if defined(INET6) || defined(INET)
1778 /* Try to queue to LRO */
1779 if (IF_LRO_ENABLED(sc) &&
1780 (cqe->u0.s.ip_cksum_pass) &&
1781 (cqe->u0.s.l4_cksum_pass) &&
1782 (!cqe->u0.s.ip_ver) &&
1783 (rq->lro.lro_cnt != 0)) {
1785 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1786 rq->lro_pkts_queued ++;
1789 /* If LRO posting fails then try to post to STACK */
1793 (*sc->ifp->if_input) (sc->ifp, m);
1794 #if defined(INET6) || defined(INET)
1797 /* Update rx stats per queue */
1798 rq->rx_stats.rx_pkts++;
1799 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1800 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1801 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1802 rq->rx_stats.rx_mcast_pkts++;
1803 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1804 rq->rx_stats.rx_ucast_pkts++;
1812 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1815 struct oce_packet_desc *pd;
1816 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1818 for (i = 0; i < num_frags; i++) {
1819 if (rq->ring->cidx == rq->ring->pidx) {
1820 device_printf(sc->dev,
1821 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1824 pd = &rq->pckts[rq->ring->cidx];
1825 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1826 bus_dmamap_unload(rq->tag, pd->map);
1827 if (pd->mbuf != NULL) {
1832 RING_GET(rq->ring, 1);
1839 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1841 struct oce_nic_rx_cqe_v1 *cqe_v1;
1844 if (sc->be3_native) {
1845 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1846 vtp = cqe_v1->u0.s.vlan_tag_present;
1848 vtp = cqe->u0.s.vlan_tag_present;
1856 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1858 struct oce_nic_rx_cqe_v1 *cqe_v1;
1861 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1862 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1863 port_id = cqe_v1->u0.s.port;
1864 if (sc->port_id != port_id)
1867 ;/* For BE3 legacy and Lancer this is dummy */
1873 #if defined(INET6) || defined(INET)
1875 oce_rx_flush_lro(struct oce_rq *rq)
1877 struct lro_ctrl *lro = &rq->lro;
1878 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1880 if (!IF_LRO_ENABLED(sc))
1883 tcp_lro_flush_all(lro);
1884 rq->lro_pkts_queued = 0;
1891 oce_init_lro(POCE_SOFTC sc)
1893 struct lro_ctrl *lro = NULL;
1896 for (i = 0; i < sc->nrqs; i++) {
1897 lro = &sc->rq[i]->lro;
1898 rc = tcp_lro_init(lro);
1900 device_printf(sc->dev, "LRO init failed\n");
1911 oce_free_lro(POCE_SOFTC sc)
1913 struct lro_ctrl *lro = NULL;
1916 for (i = 0; i < sc->nrqs; i++) {
1917 lro = &sc->rq[i]->lro;
1925 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1927 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1929 struct oce_packet_desc *pd;
1930 bus_dma_segment_t segs[6];
1931 int nsegs, added = 0;
1932 struct oce_nic_rqe *rqe;
1933 pd_rxulp_db_t rxdb_reg;
1935 uint32_t oce_max_rq_posts = 64;
1937 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1938 for (i = 0; i < count; i++) {
1939 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1941 pd = &rq->pckts[rq->ring->pidx];
1942 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1943 if (pd->mbuf == NULL) {
1944 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1947 pd->mbuf->m_nextpkt = NULL;
1949 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1951 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1954 segs, &nsegs, BUS_DMA_NOWAIT);
1957 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1966 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1968 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1969 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1970 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1971 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1972 RING_PUT(rq->ring, 1);
1976 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1978 for (i = added / oce_max_rq_posts; i > 0; i--) {
1979 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1980 rxdb_reg.bits.qid = rq->rq_id;
1982 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1983 val |= oce_max_rq_posts << 16;
1984 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1986 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1988 added -= oce_max_rq_posts;
1991 rxdb_reg.bits.qid = rq->rq_id;
1992 rxdb_reg.bits.num_posted = added;
1994 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1996 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1998 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
2007 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
2010 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
2011 if(!sc->enable_hwlro) {
2012 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
2013 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
2015 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
2016 oce_alloc_rx_bufs(rq, 64);
2024 oce_rq_handler_lro(void *arg)
2026 struct oce_rq *rq = (struct oce_rq *)arg;
2027 struct oce_cq *cq = rq->cq;
2028 POCE_SOFTC sc = rq->parent;
2029 struct nic_hwlro_singleton_cqe *cqe;
2030 struct nic_hwlro_cqe_part2 *cqe2;
2034 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2035 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2036 while (cqe->valid) {
2037 if(cqe->cqe_type == 0) { /* singleton cqe */
2038 /* we should not get singleton cqe after cqe1 on same rq */
2039 if(rq->cqe_firstpart != NULL) {
2040 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2041 goto exit_rq_handler_lro;
2043 if(cqe->error != 0) {
2044 rq->rx_stats.rxcp_err++;
2045 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2047 oce_rx_lro(rq, cqe, NULL);
2048 rq->rx_stats.rx_compl++;
2050 RING_GET(cq->ring, 1);
2052 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2054 }else if(cqe->cqe_type == 0x1) { /* first part */
2055 /* we should not get cqe1 after cqe1 on same rq */
2056 if(rq->cqe_firstpart != NULL) {
2057 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2058 goto exit_rq_handler_lro;
2060 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2061 RING_GET(cq->ring, 1);
2062 }else if(cqe->cqe_type == 0x2) { /* second part */
2063 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2064 if(cqe2->error != 0) {
2065 rq->rx_stats.rxcp_err++;
2066 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2068 /* We should not get cqe2 without cqe1 */
2069 if(rq->cqe_firstpart == NULL) {
2070 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2071 goto exit_rq_handler_lro;
2073 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2075 rq->rx_stats.rx_compl++;
2076 rq->cqe_firstpart->valid = 0;
2078 rq->cqe_firstpart = NULL;
2080 RING_GET(cq->ring, 1);
2082 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2086 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2087 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2089 oce_check_rx_bufs(sc, num_cqes, rq);
2090 exit_rq_handler_lro:
2091 UNLOCK(&rq->rx_lock);
2095 /* Handle the Completion Queue for receive */
2097 oce_rq_handler(void *arg)
2099 struct oce_rq *rq = (struct oce_rq *)arg;
2100 struct oce_cq *cq = rq->cq;
2101 POCE_SOFTC sc = rq->parent;
2102 struct oce_nic_rx_cqe *cqe;
2106 oce_rq_handler_lro(arg);
2110 bus_dmamap_sync(cq->ring->dma.tag,
2111 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2112 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2113 while (cqe->u0.dw[2]) {
2114 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2116 if (cqe->u0.s.error == 0) {
2119 rq->rx_stats.rxcp_err++;
2120 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2121 /* Post L3/L4 errors to stack.*/
2124 rq->rx_stats.rx_compl++;
2127 #if defined(INET6) || defined(INET)
2128 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2129 oce_rx_flush_lro(rq);
2133 RING_GET(cq->ring, 1);
2134 bus_dmamap_sync(cq->ring->dma.tag,
2135 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2137 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2139 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2143 #if defined(INET6) || defined(INET)
2144 if (IF_LRO_ENABLED(sc))
2145 oce_rx_flush_lro(rq);
2148 oce_check_rx_bufs(sc, num_cqes, rq);
2149 UNLOCK(&rq->rx_lock);
2157 /*****************************************************************************
2158 * Helper function prototypes in this file *
2159 *****************************************************************************/
2162 oce_attach_ifp(POCE_SOFTC sc)
2165 sc->ifp = if_alloc(IFT_ETHER);
2169 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2170 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2171 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2173 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2174 sc->ifp->if_ioctl = oce_ioctl;
2175 sc->ifp->if_start = oce_start;
2176 sc->ifp->if_init = oce_init;
2177 sc->ifp->if_mtu = ETHERMTU;
2178 sc->ifp->if_softc = sc;
2179 #if __FreeBSD_version >= 800000
2180 sc->ifp->if_transmit = oce_multiq_start;
2181 sc->ifp->if_qflush = oce_multiq_flush;
2184 if_initname(sc->ifp,
2185 device_get_name(sc->dev), device_get_unit(sc->dev));
2187 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2188 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2189 IFQ_SET_READY(&sc->ifp->if_snd);
2191 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2192 sc->ifp->if_hwassist |= CSUM_TSO;
2193 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2195 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2196 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2197 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2199 #if defined(INET6) || defined(INET)
2200 sc->ifp->if_capabilities |= IFCAP_TSO;
2201 sc->ifp->if_capabilities |= IFCAP_LRO;
2202 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2205 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2206 sc->ifp->if_baudrate = IF_Gbps(10);
2208 #if __FreeBSD_version >= 1000000
2209 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2210 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2211 sc->ifp->if_hw_tsomaxsegsize = 4096;
2214 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2221 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2223 POCE_SOFTC sc = ifp->if_softc;
2225 if (ifp->if_softc != arg)
2227 if ((vtag == 0) || (vtag > 4095))
2230 sc->vlan_tag[vtag] = 1;
2232 if (sc->vlans_added <= (sc->max_vlans + 1))
2238 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2240 POCE_SOFTC sc = ifp->if_softc;
2242 if (ifp->if_softc != arg)
2244 if ((vtag == 0) || (vtag > 4095))
2247 sc->vlan_tag[vtag] = 0;
2254 * A max of 64 vlans can be configured in BE. If the user configures
2255 * more, place the card in vlan promiscuous mode.
2258 oce_vid_config(POCE_SOFTC sc)
2260 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2261 uint16_t ntags = 0, i;
2264 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2265 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2266 for (i = 0; i < MAX_VLANS; i++) {
2267 if (sc->vlan_tag[i]) {
2268 vtags[ntags].vtag = i;
2273 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2274 vtags, ntags, 1, 0);
2276 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2283 oce_mac_addr_set(POCE_SOFTC sc)
2285 uint32_t old_pmac_id = sc->pmac_id;
2289 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2290 sc->macaddr.size_of_struct);
2294 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2295 sc->if_id, &sc->pmac_id);
2297 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2298 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2299 sc->macaddr.size_of_struct);
2302 device_printf(sc->dev, "Failed update macaddress\n");
2308 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2310 POCE_SOFTC sc = ifp->if_softc;
2311 struct ifreq *ifr = (struct ifreq *)data;
2313 char cookie[32] = {0};
2314 void *priv_data = ifr_data_get_ptr(ifr);
2318 OCE_DMA_MEM dma_mem;
2319 struct mbx_common_get_cntl_attr *fw_cmd;
2321 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2324 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2327 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2328 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2331 req_size = le32toh(req.u0.req.request_length);
2332 if (req_size > 65536)
2335 req_size += sizeof(struct mbx_hdr);
2336 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2340 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2345 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2351 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2355 firmware is filling all the attributes for this ioctl except
2356 the driver version..so fill it
2358 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2359 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2360 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2361 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2365 oce_dma_free(sc, &dma_mem);
2371 oce_eqd_set_periodic(POCE_SOFTC sc)
2373 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2374 struct oce_aic_obj *aic;
2376 uint64_t now = 0, delta;
2377 int eqd, i, num = 0;
2378 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2382 #define ticks_to_msecs(t) (1000 * (t) / hz)
2384 for (i = 0 ; i < sc->neqs; i++) {
2386 aic = &sc->aic_obj[i];
2387 /* When setting the static eq delay from the user space */
2396 rxpkts = rq->rx_stats.rx_pkts;
2398 tx_reqs = wq->tx_stats.tx_reqs;
2401 if (!aic->ticks || now < aic->ticks ||
2402 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2403 aic->prev_rxpkts = rxpkts;
2404 aic->prev_txreqs = tx_reqs;
2409 delta = ticks_to_msecs(now - aic->ticks);
2411 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2412 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2413 eqd = (pps / 15000) << 2;
2417 /* Make sure that the eq delay is in the known range */
2418 eqd = min(eqd, aic->max_eqd);
2419 eqd = max(eqd, aic->min_eqd);
2421 aic->prev_rxpkts = rxpkts;
2422 aic->prev_txreqs = tx_reqs;
2426 if (eqd != aic->cur_eqd) {
2427 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2428 set_eqd[num].eq_id = eqo->eq_id;
2434 /* Is there atleast one eq that needs to be modified? */
2435 for(i = 0; i < num; i += 8) {
2437 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2439 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2444 static void oce_detect_hw_error(POCE_SOFTC sc)
2447 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2448 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2455 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2456 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2457 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2458 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2461 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2462 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2463 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2464 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2466 ue_low = (ue_low & ~ue_low_mask);
2467 ue_high = (ue_high & ~ue_high_mask);
2470 /* On certain platforms BE hardware can indicate spurious UEs.
2471 * Allow the h/w to stop working completely in case of a real UE.
2472 * Hence not setting the hw_error for UE detection.
2474 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2475 sc->hw_error = TRUE;
2476 device_printf(sc->dev, "Error detected in the card\n");
2479 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2480 device_printf(sc->dev,
2481 "ERR: sliport status 0x%x\n", sliport_status);
2482 device_printf(sc->dev,
2483 "ERR: sliport error1 0x%x\n", sliport_err1);
2484 device_printf(sc->dev,
2485 "ERR: sliport error2 0x%x\n", sliport_err2);
2489 for (i = 0; ue_low; ue_low >>= 1, i++) {
2491 device_printf(sc->dev, "UE: %s bit set\n",
2492 ue_status_low_desc[i]);
2497 for (i = 0; ue_high; ue_high >>= 1, i++) {
2499 device_printf(sc->dev, "UE: %s bit set\n",
2500 ue_status_hi_desc[i]);
2508 oce_local_timer(void *arg)
2510 POCE_SOFTC sc = arg;
2513 oce_detect_hw_error(sc);
2514 oce_refresh_nic_stats(sc);
2515 oce_refresh_queue_stats(sc);
2516 oce_mac_addr_set(sc);
2519 for (i = 0; i < sc->nwqs; i++)
2520 oce_tx_restart(sc, sc->wq[i]);
2522 /* calculate and set the eq delay for optimal interrupt rate */
2523 if (IS_BE(sc) || IS_SH(sc))
2524 oce_eqd_set_periodic(sc);
2526 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2530 oce_tx_compl_clean(POCE_SOFTC sc)
2533 int i = 0, timeo = 0, num_wqes = 0;
2534 int pending_txqs = sc->nwqs;
2536 /* Stop polling for compls when HW has been silent for 10ms or
2537 * hw_error or no outstanding completions expected
2540 pending_txqs = sc->nwqs;
2542 for_all_wq_queues(sc, wq, i) {
2543 num_wqes = oce_wq_handler(wq);
2548 if(!wq->ring->num_used)
2552 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2558 for_all_wq_queues(sc, wq, i) {
2559 while(wq->ring->num_used) {
2560 LOCK(&wq->tx_compl_lock);
2561 oce_process_tx_completion(wq);
2562 UNLOCK(&wq->tx_compl_lock);
2568 /* NOTE : This should only be called holding
2572 oce_if_deactivate(POCE_SOFTC sc)
2579 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2581 oce_tx_compl_clean(sc);
2583 /* Stop intrs and finish any bottom halves pending */
2584 oce_hw_intr_disable(sc);
2586 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2587 any other lock. So unlock device lock and require after
2588 completing taskqueue_drain.
2590 UNLOCK(&sc->dev_lock);
2591 for (i = 0; i < sc->intr_count; i++) {
2592 if (sc->intrs[i].tq != NULL) {
2593 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2596 LOCK(&sc->dev_lock);
2598 /* Delete RX queue in card with flush param */
2601 /* Invalidate any pending cq and eq entries*/
2602 for_all_evnt_queues(sc, eq, i)
2604 for_all_rq_queues(sc, rq, i)
2605 oce_drain_rq_cq(rq);
2606 for_all_wq_queues(sc, wq, i)
2607 oce_drain_wq_cq(wq);
2609 /* But still we need to get MCC aync events.
2610 So enable intrs and also arm first EQ
2612 oce_hw_intr_enable(sc);
2613 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2620 oce_if_activate(POCE_SOFTC sc)
2627 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2629 oce_hw_intr_disable(sc);
2633 for_all_rq_queues(sc, rq, i) {
2634 rc = oce_start_rq(rq);
2636 device_printf(sc->dev, "Unable to start RX\n");
2639 for_all_wq_queues(sc, wq, i) {
2640 rc = oce_start_wq(wq);
2642 device_printf(sc->dev, "Unable to start TX\n");
2646 for_all_evnt_queues(sc, eq, i)
2647 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2649 oce_hw_intr_enable(sc);
2654 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2656 /* Update Link status */
2657 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2658 ASYNC_EVENT_LINK_UP) {
2659 sc->link_status = ASYNC_EVENT_LINK_UP;
2660 if_link_state_change(sc->ifp, LINK_STATE_UP);
2662 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2663 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2668 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2669 struct oce_async_evt_grp5_os2bmc *evt)
2671 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2672 if (evt->u.s.mgmt_enable)
2673 sc->flags |= OCE_FLAGS_OS2BMC;
2677 sc->bmc_filt_mask = evt->u.s.arp_filter;
2678 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2679 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2680 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2681 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2682 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2683 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2684 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2685 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2689 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2691 struct oce_async_event_grp5_pvid_state *gcqe;
2692 struct oce_async_evt_grp5_os2bmc *bmccqe;
2694 switch (cqe->u0.s.async_type) {
2695 case ASYNC_EVENT_PVID_STATE:
2697 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2699 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2703 case ASYNC_EVENT_OS2BMC:
2704 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2705 oce_async_grp5_osbmc_process(sc, bmccqe);
2712 /* Handle the Completion Queue for the Mailbox/Async notifications */
2714 oce_mq_handler(void *arg)
2716 struct oce_mq *mq = (struct oce_mq *)arg;
2717 POCE_SOFTC sc = mq->parent;
2718 struct oce_cq *cq = mq->cq;
2719 int num_cqes = 0, evt_type = 0, optype = 0;
2720 struct oce_mq_cqe *cqe;
2721 struct oce_async_cqe_link_state *acqe;
2722 struct oce_async_event_qnq *dbgcqe;
2725 bus_dmamap_sync(cq->ring->dma.tag,
2726 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2727 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2729 while (cqe->u0.dw[3]) {
2730 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2731 if (cqe->u0.s.async_event) {
2732 evt_type = cqe->u0.s.event_type;
2733 optype = cqe->u0.s.async_type;
2734 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2735 /* Link status evt */
2736 acqe = (struct oce_async_cqe_link_state *)cqe;
2737 process_link_state(sc, acqe);
2738 } else if (evt_type == ASYNC_EVENT_GRP5) {
2739 oce_process_grp5_events(sc, cqe);
2740 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2741 optype == ASYNC_EVENT_DEBUG_QNQ) {
2742 dbgcqe = (struct oce_async_event_qnq *)cqe;
2744 sc->qnqid = dbgcqe->vlan_tag;
2745 sc->qnq_debug_event = TRUE;
2749 RING_GET(cq->ring, 1);
2750 bus_dmamap_sync(cq->ring->dma.tag,
2751 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2752 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2757 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2764 setup_max_queues_want(POCE_SOFTC sc)
2766 /* Check if it is FLEX machine. Is so dont use RSS */
2767 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2768 (sc->function_mode & FNM_UMC_MODE) ||
2769 (sc->function_mode & FNM_VNIC_MODE) ||
2770 (!is_rss_enabled(sc)) ||
2775 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2776 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2779 if (IS_BE2(sc) && is_rss_enabled(sc))
2780 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2785 update_queues_got(POCE_SOFTC sc)
2787 if (is_rss_enabled(sc)) {
2788 sc->nrqs = sc->intr_count + 1;
2789 sc->nwqs = sc->intr_count;
2800 oce_check_ipv6_ext_hdr(struct mbuf *m)
2802 struct ether_header *eh = mtod(m, struct ether_header *);
2803 caddr_t m_datatemp = m->m_data;
2805 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2806 m->m_data += sizeof(struct ether_header);
2807 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2809 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2810 (ip6->ip6_nxt != IPPROTO_UDP)){
2811 struct ip6_ext *ip6e = NULL;
2812 m->m_data += sizeof(struct ip6_hdr);
2814 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2815 if(ip6e->ip6e_len == 0xff) {
2816 m->m_data = m_datatemp;
2820 m->m_data = m_datatemp;
2826 is_be3_a1(POCE_SOFTC sc)
2828 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2834 static struct mbuf *
2835 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2837 uint16_t vlan_tag = 0;
2842 /* Embed vlan tag in the packet if it is not part of it */
2843 if(m->m_flags & M_VLANTAG) {
2844 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2845 m->m_flags &= ~M_VLANTAG;
2848 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2851 vlan_tag = sc->pvid;
2857 m = ether_vlanencap(m, vlan_tag);
2861 m = ether_vlanencap(m, sc->qnqid);
2870 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2872 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2873 oce_check_ipv6_ext_hdr(m)) {
2880 oce_get_config(POCE_SOFTC sc)
2883 uint32_t max_rss = 0;
2885 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2886 max_rss = OCE_LEGACY_MODE_RSS;
2888 max_rss = OCE_MAX_RSS;
2891 rc = oce_get_profile_config(sc, max_rss);
2893 sc->nwqs = OCE_MAX_WQ;
2894 sc->nrssqs = max_rss;
2895 sc->nrqs = sc->nrssqs + 1;
2898 else { /* For BE3 don't rely on fw for determining the resources */
2899 sc->nrssqs = max_rss;
2900 sc->nrqs = sc->nrssqs + 1;
2901 sc->nwqs = OCE_MAX_WQ;
2902 sc->max_vlans = MAX_VLANFILTER_SIZE;
2907 oce_rdma_close(void)
2909 if (oce_rdma_if != NULL) {
2915 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2917 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2921 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2924 struct oce_dev_info di;
2927 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2931 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2932 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2936 rdma_info->close = oce_rdma_close;
2937 rdma_info->mbox_post = oce_mbox_post;
2938 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2939 rdma_info->get_mac_addr = oce_get_mac_addr;
2941 oce_rdma_if = rdma_if;
2944 while (sc != NULL) {
2945 if (oce_rdma_if->announce != NULL) {
2946 memset(&di, 0, sizeof(di));
2950 di.db_bhandle = sc->db_bhandle;
2951 di.db_btag = sc->db_btag;
2952 di.db_page_size = 4096;
2953 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2954 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2955 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2956 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2958 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2960 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2961 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2962 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2963 di.msix.start_vector = sc->intr_count;
2964 for (i=0; i<di.msix.num_vectors; i++) {
2965 di.msix.vector_list[i] = sc->intrs[i].vector;
2969 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2970 di.vendor_id = pci_get_vendor(sc->dev);
2971 di.dev_id = pci_get_device(sc->dev);
2973 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2974 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2977 rdma_if->announce(&di);
2986 oce_read_env_variables( POCE_SOFTC sc )
2991 /* read if user wants to enable hwlro or swlro */
2992 //value = getenv("oce_enable_hwlro");
2993 if(value && IS_SH(sc)) {
2994 sc->enable_hwlro = strtol(value, NULL, 10);
2995 if(sc->enable_hwlro) {
2996 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
2998 device_printf(sc->dev, "no hardware lro support\n");
2999 device_printf(sc->dev, "software lro enabled\n");
3000 sc->enable_hwlro = 0;
3002 device_printf(sc->dev, "hardware lro enabled\n");
3003 oce_max_rsp_handled = 32;
3006 device_printf(sc->dev, "software lro enabled\n");
3009 sc->enable_hwlro = 0;
3012 /* read mbuf size */
3013 //value = getenv("oce_rq_buf_size");
3014 if(value && IS_SH(sc)) {
3015 oce_rq_buf_size = strtol(value, NULL, 10);
3016 switch(oce_rq_buf_size) {
3024 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
3025 oce_rq_buf_size = 2048;