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 = oce_handle_passthrough(ifp, data);
627 rc = ether_ioctl(ifp, command, data);
642 if (sc->ifp->if_flags & IFF_UP) {
643 oce_if_deactivate(sc);
647 UNLOCK(&sc->dev_lock);
653 oce_multiq_start(struct ifnet *ifp, struct mbuf *m)
655 POCE_SOFTC sc = ifp->if_softc;
656 struct oce_wq *wq = NULL;
660 if (!sc->link_status)
663 if (M_HASHTYPE_GET(m) != M_HASHTYPE_NONE)
664 queue_index = m->m_pkthdr.flowid % sc->nwqs;
666 wq = sc->wq[queue_index];
669 status = oce_multiq_transmit(ifp, m, wq);
670 UNLOCK(&wq->tx_lock);
678 oce_multiq_flush(struct ifnet *ifp)
680 POCE_SOFTC sc = ifp->if_softc;
684 for (i = 0; i < sc->nwqs; i++) {
685 while ((m = buf_ring_dequeue_sc(sc->wq[i]->br)) != NULL)
693 /*****************************************************************************
694 * Driver interrupt routines functions *
695 *****************************************************************************/
698 oce_intr(void *arg, int pending)
701 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
702 POCE_SOFTC sc = ii->sc;
703 struct oce_eq *eq = ii->eq;
705 struct oce_cq *cq = NULL;
709 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
710 BUS_DMASYNC_POSTWRITE);
712 eqe = RING_GET_CONSUMER_ITEM_VA(eq->ring, struct oce_eqe);
716 bus_dmamap_sync(eq->ring->dma.tag, eq->ring->dma.map,
717 BUS_DMASYNC_POSTWRITE);
718 RING_GET(eq->ring, 1);
724 goto eq_arm; /* Spurious */
726 /* Clear EQ entries, but dont arm */
727 oce_arm_eq(sc, eq->eq_id, num_eqes, FALSE, FALSE);
729 /* Process TX, RX and MCC. But dont arm CQ*/
730 for (i = 0; i < eq->cq_valid; i++) {
732 (*cq->cq_handler)(cq->cb_arg);
735 /* Arm all cqs connected to this EQ */
736 for (i = 0; i < eq->cq_valid; i++) {
738 oce_arm_cq(sc, cq->cq_id, 0, TRUE);
742 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
749 oce_setup_intr(POCE_SOFTC sc)
751 int rc = 0, use_intx = 0;
752 int vector = 0, req_vectors = 0;
753 int tot_req_vectors, tot_vectors;
755 if (is_rss_enabled(sc))
756 req_vectors = MAX((sc->nrqs - 1), sc->nwqs);
760 tot_req_vectors = req_vectors;
761 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
762 if (req_vectors > 1) {
763 tot_req_vectors += OCE_RDMA_VECTORS;
764 sc->roce_intr_count = OCE_RDMA_VECTORS;
768 if (sc->flags & OCE_FLAGS_MSIX_CAPABLE) {
769 sc->intr_count = req_vectors;
770 tot_vectors = tot_req_vectors;
771 rc = pci_alloc_msix(sc->dev, &tot_vectors);
774 pci_release_msi(sc->dev);
776 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
777 if (tot_vectors < tot_req_vectors) {
778 if (sc->intr_count < (2 * OCE_RDMA_VECTORS)) {
779 sc->roce_intr_count = (tot_vectors / 2);
781 sc->intr_count = tot_vectors - sc->roce_intr_count;
784 sc->intr_count = tot_vectors;
786 sc->flags |= OCE_FLAGS_USING_MSIX;
794 /* Scale number of queues based on intr we got */
795 update_queues_got(sc);
798 device_printf(sc->dev, "Using legacy interrupt\n");
799 rc = oce_alloc_intr(sc, vector, oce_intr);
803 for (; vector < sc->intr_count; vector++) {
804 rc = oce_alloc_intr(sc, vector, oce_intr);
818 oce_fast_isr(void *arg)
820 POCE_INTR_INFO ii = (POCE_INTR_INFO) arg;
821 POCE_SOFTC sc = ii->sc;
826 oce_arm_eq(sc, ii->eq->eq_id, 0, FALSE, TRUE);
828 taskqueue_enqueue(ii->tq, &ii->task);
832 return FILTER_HANDLED;
837 oce_alloc_intr(POCE_SOFTC sc, int vector, void (*isr) (void *arg, int pending))
842 if (vector >= OCE_MAX_EQ)
845 ii = &sc->intrs[vector];
847 /* Set the resource id for the interrupt.
848 * MSIx is vector + 1 for the resource id,
849 * INTx is 0 for the resource id.
851 if (sc->flags & OCE_FLAGS_USING_MSIX)
855 ii->intr_res = bus_alloc_resource_any(sc->dev,
857 &rr, RF_ACTIVE|RF_SHAREABLE);
859 if (ii->intr_res == NULL) {
860 device_printf(sc->dev,
861 "Could not allocate interrupt\n");
866 TASK_INIT(&ii->task, 0, isr, ii);
868 sprintf(ii->task_name, "oce_task[%d]", ii->vector);
869 ii->tq = taskqueue_create_fast(ii->task_name,
871 taskqueue_thread_enqueue,
873 taskqueue_start_threads(&ii->tq, 1, PI_NET, "%s taskq",
874 device_get_nameunit(sc->dev));
877 rc = bus_setup_intr(sc->dev,
880 oce_fast_isr, NULL, ii, &ii->tag);
887 oce_intr_free(POCE_SOFTC sc)
891 for (i = 0; i < sc->intr_count; i++) {
893 if (sc->intrs[i].tag != NULL)
894 bus_teardown_intr(sc->dev, sc->intrs[i].intr_res,
896 if (sc->intrs[i].tq != NULL)
897 taskqueue_free(sc->intrs[i].tq);
899 if (sc->intrs[i].intr_res != NULL)
900 bus_release_resource(sc->dev, SYS_RES_IRQ,
902 sc->intrs[i].intr_res);
903 sc->intrs[i].tag = NULL;
904 sc->intrs[i].intr_res = NULL;
907 if (sc->flags & OCE_FLAGS_USING_MSIX)
908 pci_release_msi(sc->dev);
914 /******************************************************************************
915 * Media callbacks functions *
916 ******************************************************************************/
919 oce_media_status(struct ifnet *ifp, struct ifmediareq *req)
921 POCE_SOFTC sc = (POCE_SOFTC) ifp->if_softc;
924 req->ifm_status = IFM_AVALID;
925 req->ifm_active = IFM_ETHER;
927 if (sc->link_status == 1)
928 req->ifm_status |= IFM_ACTIVE;
932 switch (sc->link_speed) {
933 case 1: /* 10 Mbps */
934 req->ifm_active |= IFM_10_T | IFM_FDX;
937 case 2: /* 100 Mbps */
938 req->ifm_active |= IFM_100_TX | IFM_FDX;
942 req->ifm_active |= IFM_1000_T | IFM_FDX;
945 case 4: /* 10 Gbps */
946 req->ifm_active |= IFM_10G_SR | IFM_FDX;
949 case 5: /* 20 Gbps */
950 req->ifm_active |= IFM_10G_SR | IFM_FDX;
953 case 6: /* 25 Gbps */
954 req->ifm_active |= IFM_10G_SR | IFM_FDX;
957 case 7: /* 40 Gbps */
958 req->ifm_active |= IFM_40G_SR4 | IFM_FDX;
971 oce_media_change(struct ifnet *ifp)
977 static void oce_is_pkt_dest_bmc(POCE_SOFTC sc,
978 struct mbuf *m, boolean_t *os2bmc,
981 struct ether_header *eh = NULL;
983 eh = mtod(m, struct ether_header *);
985 if (!is_os2bmc_enabled(sc) || *os2bmc) {
989 if (!ETHER_IS_MULTICAST(eh->ether_dhost))
992 if (is_mc_allowed_on_bmc(sc, eh) ||
993 is_bc_allowed_on_bmc(sc, eh) ||
994 is_arp_allowed_on_bmc(sc, ntohs(eh->ether_type))) {
999 if (mtod(m, struct ip *)->ip_p == IPPROTO_IPV6) {
1000 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
1001 uint8_t nexthdr = ip6->ip6_nxt;
1002 if (nexthdr == IPPROTO_ICMPV6) {
1003 struct icmp6_hdr *icmp6 = (struct icmp6_hdr *)(ip6 + 1);
1004 switch (icmp6->icmp6_type) {
1005 case ND_ROUTER_ADVERT:
1006 *os2bmc = is_ipv6_ra_filt_enabled(sc);
1008 case ND_NEIGHBOR_ADVERT:
1009 *os2bmc = is_ipv6_na_filt_enabled(sc);
1017 if (mtod(m, struct ip *)->ip_p == IPPROTO_UDP) {
1018 struct ip *ip = mtod(m, struct ip *);
1019 int iphlen = ip->ip_hl << 2;
1020 struct udphdr *uh = (struct udphdr *)((caddr_t)ip + iphlen);
1021 switch (uh->uh_dport) {
1022 case DHCP_CLIENT_PORT:
1023 *os2bmc = is_dhcp_client_filt_enabled(sc);
1025 case DHCP_SERVER_PORT:
1026 *os2bmc = is_dhcp_srvr_filt_enabled(sc);
1028 case NET_BIOS_PORT1:
1029 case NET_BIOS_PORT2:
1030 *os2bmc = is_nbios_filt_enabled(sc);
1032 case DHCPV6_RAS_PORT:
1033 *os2bmc = is_ipv6_ras_filt_enabled(sc);
1041 *m_new = m_dup(m, M_NOWAIT);
1046 *m_new = oce_insert_vlan_tag(sc, *m_new, NULL);
1052 /*****************************************************************************
1053 * Transmit routines functions *
1054 *****************************************************************************/
1057 oce_tx(POCE_SOFTC sc, struct mbuf **mpp, int wq_index)
1059 int rc = 0, i, retry_cnt = 0;
1060 bus_dma_segment_t segs[OCE_MAX_TX_ELEMENTS];
1061 struct mbuf *m, *m_temp, *m_new = NULL;
1062 struct oce_wq *wq = sc->wq[wq_index];
1063 struct oce_packet_desc *pd;
1064 struct oce_nic_hdr_wqe *nichdr;
1065 struct oce_nic_frag_wqe *nicfrag;
1066 struct ether_header *eh = NULL;
1069 boolean_t complete = TRUE;
1070 boolean_t os2bmc = FALSE;
1076 if (!(m->m_flags & M_PKTHDR)) {
1081 /* Don't allow non-TSO packets longer than MTU */
1082 if (!is_tso_pkt(m)) {
1083 eh = mtod(m, struct ether_header *);
1084 if(m->m_pkthdr.len > ETHER_MAX_FRAME(sc->ifp, eh->ether_type, FALSE))
1088 if(oce_tx_asic_stall_verify(sc, m)) {
1089 m = oce_insert_vlan_tag(sc, m, &complete);
1091 device_printf(sc->dev, "Insertion unsuccessful\n");
1097 /* Lancer, SH ASIC has a bug wherein Packets that are 32 bytes or less
1098 * may cause a transmit stall on that port. So the work-around is to
1099 * pad short packets (<= 32 bytes) to a 36-byte length.
1101 if(IS_SH(sc) || IS_XE201(sc) ) {
1102 if(m->m_pkthdr.len <= 32) {
1104 bzero((void *)buf, 36);
1105 m_append(m, (36 - m->m_pkthdr.len), buf);
1110 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1111 /* consolidate packet buffers for TSO/LSO segment offload */
1112 #if defined(INET6) || defined(INET)
1113 m = oce_tso_setup(sc, mpp);
1124 pd = &wq->pckts[wq->pkt_desc_head];
1127 rc = bus_dmamap_load_mbuf_sg(wq->tag,
1129 m, segs, &pd->nsegs, BUS_DMA_NOWAIT);
1131 num_wqes = pd->nsegs + 1;
1132 if (IS_BE(sc) || IS_SH(sc)) {
1133 /*Dummy required only for BE3.*/
1137 if (num_wqes >= RING_NUM_FREE(wq->ring)) {
1138 bus_dmamap_unload(wq->tag, pd->map);
1141 atomic_store_rel_int(&wq->pkt_desc_head,
1142 (wq->pkt_desc_head + 1) % \
1143 OCE_WQ_PACKET_ARRAY_SIZE);
1144 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_PREWRITE);
1148 RING_GET_PRODUCER_ITEM_VA(wq->ring, struct oce_nic_hdr_wqe);
1149 nichdr->u0.dw[0] = 0;
1150 nichdr->u0.dw[1] = 0;
1151 nichdr->u0.dw[2] = 0;
1152 nichdr->u0.dw[3] = 0;
1154 nichdr->u0.s.complete = complete;
1155 nichdr->u0.s.mgmt = os2bmc;
1156 nichdr->u0.s.event = 1;
1157 nichdr->u0.s.crc = 1;
1158 nichdr->u0.s.forward = 0;
1159 nichdr->u0.s.ipcs = (m->m_pkthdr.csum_flags & CSUM_IP) ? 1 : 0;
1160 nichdr->u0.s.udpcs =
1161 (m->m_pkthdr.csum_flags & CSUM_UDP) ? 1 : 0;
1162 nichdr->u0.s.tcpcs =
1163 (m->m_pkthdr.csum_flags & CSUM_TCP) ? 1 : 0;
1164 nichdr->u0.s.num_wqe = num_wqes;
1165 nichdr->u0.s.total_length = m->m_pkthdr.len;
1167 if (m->m_flags & M_VLANTAG) {
1168 nichdr->u0.s.vlan = 1; /*Vlan present*/
1169 nichdr->u0.s.vlan_tag = m->m_pkthdr.ether_vtag;
1172 if (m->m_pkthdr.csum_flags & CSUM_TSO) {
1173 if (m->m_pkthdr.tso_segsz) {
1174 nichdr->u0.s.lso = 1;
1175 nichdr->u0.s.lso_mss = m->m_pkthdr.tso_segsz;
1177 if (!IS_BE(sc) || !IS_SH(sc))
1178 nichdr->u0.s.ipcs = 1;
1181 RING_PUT(wq->ring, 1);
1182 atomic_add_int(&wq->ring->num_used, 1);
1184 for (i = 0; i < pd->nsegs; i++) {
1186 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1187 struct oce_nic_frag_wqe);
1188 nicfrag->u0.s.rsvd0 = 0;
1189 nicfrag->u0.s.frag_pa_hi = ADDR_HI(segs[i].ds_addr);
1190 nicfrag->u0.s.frag_pa_lo = ADDR_LO(segs[i].ds_addr);
1191 nicfrag->u0.s.frag_len = segs[i].ds_len;
1192 pd->wqe_idx = wq->ring->pidx;
1193 RING_PUT(wq->ring, 1);
1194 atomic_add_int(&wq->ring->num_used, 1);
1196 if (num_wqes > (pd->nsegs + 1)) {
1198 RING_GET_PRODUCER_ITEM_VA(wq->ring,
1199 struct oce_nic_frag_wqe);
1200 nicfrag->u0.dw[0] = 0;
1201 nicfrag->u0.dw[1] = 0;
1202 nicfrag->u0.dw[2] = 0;
1203 nicfrag->u0.dw[3] = 0;
1204 pd->wqe_idx = wq->ring->pidx;
1205 RING_PUT(wq->ring, 1);
1206 atomic_add_int(&wq->ring->num_used, 1);
1210 if_inc_counter(sc->ifp, IFCOUNTER_OPACKETS, 1);
1211 wq->tx_stats.tx_reqs++;
1212 wq->tx_stats.tx_wrbs += num_wqes;
1213 wq->tx_stats.tx_bytes += m->m_pkthdr.len;
1214 wq->tx_stats.tx_pkts++;
1216 bus_dmamap_sync(wq->ring->dma.tag, wq->ring->dma.map,
1217 BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE);
1218 reg_value = (num_wqes << 16) | wq->wq_id;
1220 /* if os2bmc is not enabled or if the pkt is already tagged as
1223 oce_is_pkt_dest_bmc(sc, m, &os2bmc, &m_new);
1225 OCE_WRITE_REG32(sc, db, wq->db_offset, reg_value);
1227 } else if (rc == EFBIG) {
1228 if (retry_cnt == 0) {
1229 m_temp = m_defrag(m, M_NOWAIT);
1234 retry_cnt = retry_cnt + 1;
1238 } else if (rc == ENOMEM)
1258 oce_process_tx_completion(struct oce_wq *wq)
1260 struct oce_packet_desc *pd;
1261 POCE_SOFTC sc = (POCE_SOFTC) wq->parent;
1264 pd = &wq->pckts[wq->pkt_desc_tail];
1265 atomic_store_rel_int(&wq->pkt_desc_tail,
1266 (wq->pkt_desc_tail + 1) % OCE_WQ_PACKET_ARRAY_SIZE);
1267 atomic_subtract_int(&wq->ring->num_used, pd->nsegs + 1);
1268 bus_dmamap_sync(wq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1269 bus_dmamap_unload(wq->tag, pd->map);
1276 if (sc->ifp->if_drv_flags & IFF_DRV_OACTIVE) {
1277 if (wq->ring->num_used < (wq->ring->num_items / 2)) {
1278 sc->ifp->if_drv_flags &= ~(IFF_DRV_OACTIVE);
1279 oce_tx_restart(sc, wq);
1286 oce_tx_restart(POCE_SOFTC sc, struct oce_wq *wq)
1289 if ((sc->ifp->if_drv_flags & IFF_DRV_RUNNING) != IFF_DRV_RUNNING)
1292 #if __FreeBSD_version >= 800000
1293 if (!drbr_empty(sc->ifp, wq->br))
1295 if (!IFQ_DRV_IS_EMPTY(&sc->ifp->if_snd))
1297 taskqueue_enqueue(taskqueue_swi, &wq->txtask);
1302 #if defined(INET6) || defined(INET)
1303 static struct mbuf *
1304 oce_tso_setup(POCE_SOFTC sc, struct mbuf **mpp)
1311 struct ip6_hdr *ip6;
1313 struct ether_vlan_header *eh;
1316 int total_len = 0, ehdrlen = 0;
1320 if (M_WRITABLE(m) == 0) {
1321 m = m_dup(*mpp, M_NOWAIT);
1328 eh = mtod(m, struct ether_vlan_header *);
1329 if (eh->evl_encap_proto == htons(ETHERTYPE_VLAN)) {
1330 etype = ntohs(eh->evl_proto);
1331 ehdrlen = ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN;
1333 etype = ntohs(eh->evl_encap_proto);
1334 ehdrlen = ETHER_HDR_LEN;
1340 ip = (struct ip *)(m->m_data + ehdrlen);
1341 if (ip->ip_p != IPPROTO_TCP)
1343 th = (struct tcphdr *)((caddr_t)ip + (ip->ip_hl << 2));
1345 total_len = ehdrlen + (ip->ip_hl << 2) + (th->th_off << 2);
1349 case ETHERTYPE_IPV6:
1350 ip6 = (struct ip6_hdr *)(m->m_data + ehdrlen);
1351 if (ip6->ip6_nxt != IPPROTO_TCP)
1353 th = (struct tcphdr *)((caddr_t)ip6 + sizeof(struct ip6_hdr));
1355 total_len = ehdrlen + sizeof(struct ip6_hdr) + (th->th_off << 2);
1362 m = m_pullup(m, total_len);
1369 #endif /* INET6 || INET */
1372 oce_tx_task(void *arg, int npending)
1374 struct oce_wq *wq = arg;
1375 POCE_SOFTC sc = wq->parent;
1376 struct ifnet *ifp = sc->ifp;
1379 #if __FreeBSD_version >= 800000
1381 rc = oce_multiq_transmit(ifp, NULL, wq);
1383 device_printf(sc->dev,
1384 "TX[%d] restart failed\n", wq->queue_index);
1386 UNLOCK(&wq->tx_lock);
1395 oce_start(struct ifnet *ifp)
1397 POCE_SOFTC sc = ifp->if_softc;
1400 int def_q = 0; /* Defualt tx queue is 0*/
1402 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1406 if (!sc->link_status)
1410 IF_DEQUEUE(&sc->ifp->if_snd, m);
1414 LOCK(&sc->wq[def_q]->tx_lock);
1415 rc = oce_tx(sc, &m, def_q);
1416 UNLOCK(&sc->wq[def_q]->tx_lock);
1419 sc->wq[def_q]->tx_stats.tx_stops ++;
1420 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1421 IFQ_DRV_PREPEND(&ifp->if_snd, m);
1427 ETHER_BPF_MTAP(ifp, m);
1435 /* Handle the Completion Queue for transmit */
1437 oce_wq_handler(void *arg)
1439 struct oce_wq *wq = (struct oce_wq *)arg;
1440 POCE_SOFTC sc = wq->parent;
1441 struct oce_cq *cq = wq->cq;
1442 struct oce_nic_tx_cqe *cqe;
1445 LOCK(&wq->tx_compl_lock);
1446 bus_dmamap_sync(cq->ring->dma.tag,
1447 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1448 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1449 while (cqe->u0.dw[3]) {
1450 DW_SWAP((uint32_t *) cqe, sizeof(oce_wq_cqe));
1452 wq->ring->cidx = cqe->u0.s.wqe_index + 1;
1453 if (wq->ring->cidx >= wq->ring->num_items)
1454 wq->ring->cidx -= wq->ring->num_items;
1456 oce_process_tx_completion(wq);
1457 wq->tx_stats.tx_compl++;
1459 RING_GET(cq->ring, 1);
1460 bus_dmamap_sync(cq->ring->dma.tag,
1461 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
1463 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_tx_cqe);
1468 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
1470 UNLOCK(&wq->tx_compl_lock);
1476 oce_multiq_transmit(struct ifnet *ifp, struct mbuf *m, struct oce_wq *wq)
1478 POCE_SOFTC sc = ifp->if_softc;
1479 int status = 0, queue_index = 0;
1480 struct mbuf *next = NULL;
1481 struct buf_ring *br = NULL;
1484 queue_index = wq->queue_index;
1486 if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) !=
1489 status = drbr_enqueue(ifp, br, m);
1494 if ((status = drbr_enqueue(ifp, br, m)) != 0)
1497 while ((next = drbr_peek(ifp, br)) != NULL) {
1498 if (oce_tx(sc, &next, queue_index)) {
1500 drbr_advance(ifp, br);
1502 drbr_putback(ifp, br, next);
1503 wq->tx_stats.tx_stops ++;
1504 ifp->if_drv_flags |= IFF_DRV_OACTIVE;
1508 drbr_advance(ifp, br);
1509 if_inc_counter(ifp, IFCOUNTER_OBYTES, next->m_pkthdr.len);
1510 if (next->m_flags & M_MCAST)
1511 if_inc_counter(ifp, IFCOUNTER_OMCASTS, 1);
1512 ETHER_BPF_MTAP(ifp, next);
1521 /*****************************************************************************
1522 * Receive routines functions *
1523 *****************************************************************************/
1526 oce_correct_header(struct mbuf *m, struct nic_hwlro_cqe_part1 *cqe1, struct nic_hwlro_cqe_part2 *cqe2)
1529 struct ether_header *eh = NULL;
1530 struct tcphdr *tcp_hdr = NULL;
1531 struct ip *ip4_hdr = NULL;
1532 struct ip6_hdr *ip6 = NULL;
1533 uint32_t payload_len = 0;
1535 eh = mtod(m, struct ether_header *);
1536 /* correct IP header */
1537 if(!cqe2->ipv6_frame) {
1538 ip4_hdr = (struct ip *)((char*)eh + sizeof(struct ether_header));
1539 ip4_hdr->ip_ttl = cqe2->frame_lifespan;
1540 ip4_hdr->ip_len = htons(cqe2->coalesced_size - sizeof(struct ether_header));
1541 tcp_hdr = (struct tcphdr *)((char*)ip4_hdr + sizeof(struct ip));
1543 ip6 = (struct ip6_hdr *)((char*)eh + sizeof(struct ether_header));
1544 ip6->ip6_ctlun.ip6_un1.ip6_un1_hlim = cqe2->frame_lifespan;
1545 payload_len = cqe2->coalesced_size - sizeof(struct ether_header)
1546 - sizeof(struct ip6_hdr);
1547 ip6->ip6_ctlun.ip6_un1.ip6_un1_plen = htons(payload_len);
1548 tcp_hdr = (struct tcphdr *)((char*)ip6 + sizeof(struct ip6_hdr));
1551 /* correct tcp header */
1552 tcp_hdr->th_ack = htonl(cqe2->tcp_ack_num);
1554 tcp_hdr->th_flags |= TH_PUSH;
1556 tcp_hdr->th_win = htons(cqe2->tcp_window);
1557 tcp_hdr->th_sum = 0xffff;
1559 p = (uint32_t *)((char*)tcp_hdr + sizeof(struct tcphdr) + 2);
1560 *p = cqe1->tcp_timestamp_val;
1561 *(p+1) = cqe1->tcp_timestamp_ecr;
1568 oce_rx_mbuf_chain(struct oce_rq *rq, struct oce_common_cqe_info *cqe_info, struct mbuf **m)
1570 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1571 uint32_t i = 0, frag_len = 0;
1572 uint32_t len = cqe_info->pkt_size;
1573 struct oce_packet_desc *pd;
1574 struct mbuf *tail = NULL;
1576 for (i = 0; i < cqe_info->num_frags; i++) {
1577 if (rq->ring->cidx == rq->ring->pidx) {
1578 device_printf(sc->dev,
1579 "oce_rx_mbuf_chain: Invalid RX completion - Queue is empty\n");
1582 pd = &rq->pckts[rq->ring->cidx];
1584 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1585 bus_dmamap_unload(rq->tag, pd->map);
1586 RING_GET(rq->ring, 1);
1589 frag_len = (len > rq->cfg.frag_size) ? rq->cfg.frag_size : len;
1590 pd->mbuf->m_len = frag_len;
1593 /* additional fragments */
1594 pd->mbuf->m_flags &= ~M_PKTHDR;
1595 tail->m_next = pd->mbuf;
1597 tail->m_nextpkt = NULL;
1600 /* first fragment, fill out much of the packet header */
1601 pd->mbuf->m_pkthdr.len = len;
1603 pd->mbuf->m_nextpkt = NULL;
1604 pd->mbuf->m_pkthdr.csum_flags = 0;
1605 if (IF_CSUM_ENABLED(sc)) {
1606 if (cqe_info->l4_cksum_pass) {
1607 if(!cqe_info->ipv6_frame) { /* IPV4 */
1608 pd->mbuf->m_pkthdr.csum_flags |=
1609 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1610 }else { /* IPV6 frame */
1612 pd->mbuf->m_pkthdr.csum_flags |=
1613 (CSUM_DATA_VALID | CSUM_PSEUDO_HDR);
1616 pd->mbuf->m_pkthdr.csum_data = 0xffff;
1618 if (cqe_info->ip_cksum_pass) {
1619 pd->mbuf->m_pkthdr.csum_flags |=
1620 (CSUM_IP_CHECKED|CSUM_IP_VALID);
1623 *m = tail = pd->mbuf;
1633 oce_rx_lro(struct oce_rq *rq, struct nic_hwlro_singleton_cqe *cqe, struct nic_hwlro_cqe_part2 *cqe2)
1635 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1636 struct nic_hwlro_cqe_part1 *cqe1 = NULL;
1637 struct mbuf *m = NULL;
1638 struct oce_common_cqe_info cq_info;
1642 cq_info.pkt_size = cqe->pkt_size;
1643 cq_info.vtag = cqe->vlan_tag;
1644 cq_info.l4_cksum_pass = cqe->l4_cksum_pass;
1645 cq_info.ip_cksum_pass = cqe->ip_cksum_pass;
1646 cq_info.ipv6_frame = cqe->ipv6_frame;
1647 cq_info.vtp = cqe->vtp;
1648 cq_info.qnq = cqe->qnq;
1650 cqe1 = (struct nic_hwlro_cqe_part1 *)cqe;
1651 cq_info.pkt_size = cqe2->coalesced_size;
1652 cq_info.vtag = cqe2->vlan_tag;
1653 cq_info.l4_cksum_pass = cqe2->l4_cksum_pass;
1654 cq_info.ip_cksum_pass = cqe2->ip_cksum_pass;
1655 cq_info.ipv6_frame = cqe2->ipv6_frame;
1656 cq_info.vtp = cqe2->vtp;
1657 cq_info.qnq = cqe1->qnq;
1660 cq_info.vtag = BSWAP_16(cq_info.vtag);
1662 cq_info.num_frags = cq_info.pkt_size / rq->cfg.frag_size;
1663 if(cq_info.pkt_size % rq->cfg.frag_size)
1664 cq_info.num_frags++;
1666 oce_rx_mbuf_chain(rq, &cq_info, &m);
1670 //assert(cqe2->valid != 0);
1672 //assert(cqe2->cqe_type != 2);
1673 oce_correct_header(m, cqe1, cqe2);
1676 m->m_pkthdr.rcvif = sc->ifp;
1677 #if __FreeBSD_version >= 800000
1678 if (rq->queue_index)
1679 m->m_pkthdr.flowid = (rq->queue_index - 1);
1681 m->m_pkthdr.flowid = rq->queue_index;
1682 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1684 /* This deternies if vlan tag is Valid */
1686 if (sc->function_mode & FNM_FLEX10_MODE) {
1687 /* FLEX10. If QnQ is not set, neglect VLAN */
1689 m->m_pkthdr.ether_vtag = cq_info.vtag;
1690 m->m_flags |= M_VLANTAG;
1692 } else if (sc->pvid != (cq_info.vtag & VLAN_VID_MASK)) {
1693 /* In UMC mode generally pvid will be striped by
1694 hw. But in some cases we have seen it comes
1695 with pvid. So if pvid == vlan, neglect vlan.
1697 m->m_pkthdr.ether_vtag = cq_info.vtag;
1698 m->m_flags |= M_VLANTAG;
1701 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1703 (*sc->ifp->if_input) (sc->ifp, m);
1705 /* Update rx stats per queue */
1706 rq->rx_stats.rx_pkts++;
1707 rq->rx_stats.rx_bytes += cq_info.pkt_size;
1708 rq->rx_stats.rx_frags += cq_info.num_frags;
1709 rq->rx_stats.rx_ucast_pkts++;
1715 oce_rx(struct oce_rq *rq, struct oce_nic_rx_cqe *cqe)
1717 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1719 struct mbuf *m = NULL;
1720 struct oce_common_cqe_info cq_info;
1723 /* Is it a flush compl that has no data */
1724 if(!cqe->u0.s.num_fragments)
1727 len = cqe->u0.s.pkt_size;
1729 /*partial DMA workaround for Lancer*/
1730 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1734 if (!oce_cqe_portid_valid(sc, cqe)) {
1735 oce_discard_rx_comp(rq, cqe->u0.s.num_fragments);
1739 /* Get vlan_tag value */
1740 if(IS_BE(sc) || IS_SH(sc))
1741 vtag = BSWAP_16(cqe->u0.s.vlan_tag);
1743 vtag = cqe->u0.s.vlan_tag;
1745 cq_info.l4_cksum_pass = cqe->u0.s.l4_cksum_pass;
1746 cq_info.ip_cksum_pass = cqe->u0.s.ip_cksum_pass;
1747 cq_info.ipv6_frame = cqe->u0.s.ip_ver;
1748 cq_info.num_frags = cqe->u0.s.num_fragments;
1749 cq_info.pkt_size = cqe->u0.s.pkt_size;
1751 oce_rx_mbuf_chain(rq, &cq_info, &m);
1754 m->m_pkthdr.rcvif = sc->ifp;
1755 #if __FreeBSD_version >= 800000
1756 if (rq->queue_index)
1757 m->m_pkthdr.flowid = (rq->queue_index - 1);
1759 m->m_pkthdr.flowid = rq->queue_index;
1760 M_HASHTYPE_SET(m, M_HASHTYPE_OPAQUE);
1762 /* This deternies if vlan tag is Valid */
1763 if (oce_cqe_vtp_valid(sc, cqe)) {
1764 if (sc->function_mode & FNM_FLEX10_MODE) {
1765 /* FLEX10. If QnQ is not set, neglect VLAN */
1766 if (cqe->u0.s.qnq) {
1767 m->m_pkthdr.ether_vtag = vtag;
1768 m->m_flags |= M_VLANTAG;
1770 } else if (sc->pvid != (vtag & VLAN_VID_MASK)) {
1771 /* In UMC mode generally pvid will be striped by
1772 hw. But in some cases we have seen it comes
1773 with pvid. So if pvid == vlan, neglect vlan.
1775 m->m_pkthdr.ether_vtag = vtag;
1776 m->m_flags |= M_VLANTAG;
1780 if_inc_counter(sc->ifp, IFCOUNTER_IPACKETS, 1);
1781 #if defined(INET6) || defined(INET)
1782 /* Try to queue to LRO */
1783 if (IF_LRO_ENABLED(sc) &&
1784 (cqe->u0.s.ip_cksum_pass) &&
1785 (cqe->u0.s.l4_cksum_pass) &&
1786 (!cqe->u0.s.ip_ver) &&
1787 (rq->lro.lro_cnt != 0)) {
1789 if (tcp_lro_rx(&rq->lro, m, 0) == 0) {
1790 rq->lro_pkts_queued ++;
1793 /* If LRO posting fails then try to post to STACK */
1797 (*sc->ifp->if_input) (sc->ifp, m);
1798 #if defined(INET6) || defined(INET)
1801 /* Update rx stats per queue */
1802 rq->rx_stats.rx_pkts++;
1803 rq->rx_stats.rx_bytes += cqe->u0.s.pkt_size;
1804 rq->rx_stats.rx_frags += cqe->u0.s.num_fragments;
1805 if (cqe->u0.s.pkt_type == OCE_MULTICAST_PACKET)
1806 rq->rx_stats.rx_mcast_pkts++;
1807 if (cqe->u0.s.pkt_type == OCE_UNICAST_PACKET)
1808 rq->rx_stats.rx_ucast_pkts++;
1816 oce_discard_rx_comp(struct oce_rq *rq, int num_frags)
1819 struct oce_packet_desc *pd;
1820 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1822 for (i = 0; i < num_frags; i++) {
1823 if (rq->ring->cidx == rq->ring->pidx) {
1824 device_printf(sc->dev,
1825 "oce_discard_rx_comp: Invalid RX completion - Queue is empty\n");
1828 pd = &rq->pckts[rq->ring->cidx];
1829 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_POSTWRITE);
1830 bus_dmamap_unload(rq->tag, pd->map);
1831 if (pd->mbuf != NULL) {
1836 RING_GET(rq->ring, 1);
1843 oce_cqe_vtp_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1845 struct oce_nic_rx_cqe_v1 *cqe_v1;
1848 if (sc->be3_native) {
1849 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1850 vtp = cqe_v1->u0.s.vlan_tag_present;
1852 vtp = cqe->u0.s.vlan_tag_present;
1860 oce_cqe_portid_valid(POCE_SOFTC sc, struct oce_nic_rx_cqe *cqe)
1862 struct oce_nic_rx_cqe_v1 *cqe_v1;
1865 if (sc->be3_native && (IS_BE(sc) || IS_SH(sc))) {
1866 cqe_v1 = (struct oce_nic_rx_cqe_v1 *)cqe;
1867 port_id = cqe_v1->u0.s.port;
1868 if (sc->port_id != port_id)
1871 ;/* For BE3 legacy and Lancer this is dummy */
1877 #if defined(INET6) || defined(INET)
1879 oce_rx_flush_lro(struct oce_rq *rq)
1881 struct lro_ctrl *lro = &rq->lro;
1882 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1884 if (!IF_LRO_ENABLED(sc))
1887 tcp_lro_flush_all(lro);
1888 rq->lro_pkts_queued = 0;
1895 oce_init_lro(POCE_SOFTC sc)
1897 struct lro_ctrl *lro = NULL;
1900 for (i = 0; i < sc->nrqs; i++) {
1901 lro = &sc->rq[i]->lro;
1902 rc = tcp_lro_init(lro);
1904 device_printf(sc->dev, "LRO init failed\n");
1915 oce_free_lro(POCE_SOFTC sc)
1917 struct lro_ctrl *lro = NULL;
1920 for (i = 0; i < sc->nrqs; i++) {
1921 lro = &sc->rq[i]->lro;
1929 oce_alloc_rx_bufs(struct oce_rq *rq, int count)
1931 POCE_SOFTC sc = (POCE_SOFTC) rq->parent;
1933 struct oce_packet_desc *pd;
1934 bus_dma_segment_t segs[6];
1935 int nsegs, added = 0;
1936 struct oce_nic_rqe *rqe;
1937 pd_rxulp_db_t rxdb_reg;
1939 uint32_t oce_max_rq_posts = 64;
1941 bzero(&rxdb_reg, sizeof(pd_rxulp_db_t));
1942 for (i = 0; i < count; i++) {
1943 in = (rq->ring->pidx + 1) % OCE_RQ_PACKET_ARRAY_SIZE;
1945 pd = &rq->pckts[rq->ring->pidx];
1946 pd->mbuf = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, oce_rq_buf_size);
1947 if (pd->mbuf == NULL) {
1948 device_printf(sc->dev, "mbuf allocation failed, size = %d\n",oce_rq_buf_size);
1951 pd->mbuf->m_nextpkt = NULL;
1953 pd->mbuf->m_len = pd->mbuf->m_pkthdr.len = rq->cfg.frag_size;
1955 rc = bus_dmamap_load_mbuf_sg(rq->tag,
1958 segs, &nsegs, BUS_DMA_NOWAIT);
1961 device_printf(sc->dev, "bus_dmamap_load_mbuf_sg failed rc = %d\n", rc);
1970 bus_dmamap_sync(rq->tag, pd->map, BUS_DMASYNC_PREREAD);
1972 rqe = RING_GET_PRODUCER_ITEM_VA(rq->ring, struct oce_nic_rqe);
1973 rqe->u0.s.frag_pa_hi = ADDR_HI(segs[0].ds_addr);
1974 rqe->u0.s.frag_pa_lo = ADDR_LO(segs[0].ds_addr);
1975 DW_SWAP(u32ptr(rqe), sizeof(struct oce_nic_rqe));
1976 RING_PUT(rq->ring, 1);
1980 oce_max_rq_posts = sc->enable_hwlro ? OCE_HWLRO_MAX_RQ_POSTS : OCE_MAX_RQ_POSTS;
1982 for (i = added / oce_max_rq_posts; i > 0; i--) {
1983 rxdb_reg.bits.num_posted = oce_max_rq_posts;
1984 rxdb_reg.bits.qid = rq->rq_id;
1986 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
1987 val |= oce_max_rq_posts << 16;
1988 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
1990 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
1992 added -= oce_max_rq_posts;
1995 rxdb_reg.bits.qid = rq->rq_id;
1996 rxdb_reg.bits.num_posted = added;
1998 val |= rq->rq_id & DB_LRO_RQ_ID_MASK;
2000 OCE_WRITE_REG32(sc, db, DB_OFFSET, val);
2002 OCE_WRITE_REG32(sc, db, PD_RXULP_DB, rxdb_reg.dw0);
2011 oce_check_rx_bufs(POCE_SOFTC sc, uint32_t num_cqes, struct oce_rq *rq)
2014 oce_arm_cq(sc, rq->cq->cq_id, num_cqes, FALSE);
2015 if(!sc->enable_hwlro) {
2016 if((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) > 1)
2017 oce_alloc_rx_bufs(rq, ((OCE_RQ_PACKET_ARRAY_SIZE - rq->pending) - 1));
2019 if ((OCE_RQ_PACKET_ARRAY_SIZE -1 - rq->pending) > 64)
2020 oce_alloc_rx_bufs(rq, 64);
2028 oce_rq_handler_lro(void *arg)
2030 struct oce_rq *rq = (struct oce_rq *)arg;
2031 struct oce_cq *cq = rq->cq;
2032 POCE_SOFTC sc = rq->parent;
2033 struct nic_hwlro_singleton_cqe *cqe;
2034 struct nic_hwlro_cqe_part2 *cqe2;
2038 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2039 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2040 while (cqe->valid) {
2041 if(cqe->cqe_type == 0) { /* singleton cqe */
2042 /* we should not get singleton cqe after cqe1 on same rq */
2043 if(rq->cqe_firstpart != NULL) {
2044 device_printf(sc->dev, "Got singleton cqe after cqe1 \n");
2045 goto exit_rq_handler_lro;
2047 if(cqe->error != 0) {
2048 rq->rx_stats.rxcp_err++;
2049 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2051 oce_rx_lro(rq, cqe, NULL);
2052 rq->rx_stats.rx_compl++;
2054 RING_GET(cq->ring, 1);
2056 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2058 }else if(cqe->cqe_type == 0x1) { /* first part */
2059 /* we should not get cqe1 after cqe1 on same rq */
2060 if(rq->cqe_firstpart != NULL) {
2061 device_printf(sc->dev, "Got cqe1 after cqe1 \n");
2062 goto exit_rq_handler_lro;
2064 rq->cqe_firstpart = (struct nic_hwlro_cqe_part1 *)cqe;
2065 RING_GET(cq->ring, 1);
2066 }else if(cqe->cqe_type == 0x2) { /* second part */
2067 cqe2 = (struct nic_hwlro_cqe_part2 *)cqe;
2068 if(cqe2->error != 0) {
2069 rq->rx_stats.rxcp_err++;
2070 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2072 /* We should not get cqe2 without cqe1 */
2073 if(rq->cqe_firstpart == NULL) {
2074 device_printf(sc->dev, "Got cqe2 without cqe1 \n");
2075 goto exit_rq_handler_lro;
2077 oce_rx_lro(rq, (struct nic_hwlro_singleton_cqe *)rq->cqe_firstpart, cqe2);
2079 rq->rx_stats.rx_compl++;
2080 rq->cqe_firstpart->valid = 0;
2082 rq->cqe_firstpart = NULL;
2084 RING_GET(cq->ring, 1);
2086 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2090 bus_dmamap_sync(cq->ring->dma.tag,cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2091 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct nic_hwlro_singleton_cqe);
2093 oce_check_rx_bufs(sc, num_cqes, rq);
2094 exit_rq_handler_lro:
2095 UNLOCK(&rq->rx_lock);
2099 /* Handle the Completion Queue for receive */
2101 oce_rq_handler(void *arg)
2103 struct oce_rq *rq = (struct oce_rq *)arg;
2104 struct oce_cq *cq = rq->cq;
2105 POCE_SOFTC sc = rq->parent;
2106 struct oce_nic_rx_cqe *cqe;
2110 oce_rq_handler_lro(arg);
2114 bus_dmamap_sync(cq->ring->dma.tag,
2115 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2116 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2117 while (cqe->u0.dw[2]) {
2118 DW_SWAP((uint32_t *) cqe, sizeof(oce_rq_cqe));
2120 if (cqe->u0.s.error == 0) {
2123 rq->rx_stats.rxcp_err++;
2124 if_inc_counter(sc->ifp, IFCOUNTER_IERRORS, 1);
2125 /* Post L3/L4 errors to stack.*/
2128 rq->rx_stats.rx_compl++;
2131 #if defined(INET6) || defined(INET)
2132 if (IF_LRO_ENABLED(sc) && rq->lro_pkts_queued >= 16) {
2133 oce_rx_flush_lro(rq);
2137 RING_GET(cq->ring, 1);
2138 bus_dmamap_sync(cq->ring->dma.tag,
2139 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2141 RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_nic_rx_cqe);
2143 if (num_cqes >= (IS_XE201(sc) ? 8 : oce_max_rsp_handled))
2147 #if defined(INET6) || defined(INET)
2148 if (IF_LRO_ENABLED(sc))
2149 oce_rx_flush_lro(rq);
2152 oce_check_rx_bufs(sc, num_cqes, rq);
2153 UNLOCK(&rq->rx_lock);
2161 /*****************************************************************************
2162 * Helper function prototypes in this file *
2163 *****************************************************************************/
2166 oce_attach_ifp(POCE_SOFTC sc)
2169 sc->ifp = if_alloc(IFT_ETHER);
2173 ifmedia_init(&sc->media, IFM_IMASK, oce_media_change, oce_media_status);
2174 ifmedia_add(&sc->media, IFM_ETHER | IFM_AUTO, 0, NULL);
2175 ifmedia_set(&sc->media, IFM_ETHER | IFM_AUTO);
2177 sc->ifp->if_flags = IFF_BROADCAST | IFF_MULTICAST;
2178 sc->ifp->if_ioctl = oce_ioctl;
2179 sc->ifp->if_start = oce_start;
2180 sc->ifp->if_init = oce_init;
2181 sc->ifp->if_mtu = ETHERMTU;
2182 sc->ifp->if_softc = sc;
2183 #if __FreeBSD_version >= 800000
2184 sc->ifp->if_transmit = oce_multiq_start;
2185 sc->ifp->if_qflush = oce_multiq_flush;
2188 if_initname(sc->ifp,
2189 device_get_name(sc->dev), device_get_unit(sc->dev));
2191 sc->ifp->if_snd.ifq_drv_maxlen = OCE_MAX_TX_DESC - 1;
2192 IFQ_SET_MAXLEN(&sc->ifp->if_snd, sc->ifp->if_snd.ifq_drv_maxlen);
2193 IFQ_SET_READY(&sc->ifp->if_snd);
2195 sc->ifp->if_hwassist = OCE_IF_HWASSIST;
2196 sc->ifp->if_hwassist |= CSUM_TSO;
2197 sc->ifp->if_hwassist |= (CSUM_IP | CSUM_TCP | CSUM_UDP);
2199 sc->ifp->if_capabilities = OCE_IF_CAPABILITIES;
2200 sc->ifp->if_capabilities |= IFCAP_HWCSUM;
2201 sc->ifp->if_capabilities |= IFCAP_VLAN_HWFILTER;
2203 #if defined(INET6) || defined(INET)
2204 sc->ifp->if_capabilities |= IFCAP_TSO;
2205 sc->ifp->if_capabilities |= IFCAP_LRO;
2206 sc->ifp->if_capabilities |= IFCAP_VLAN_HWTSO;
2209 sc->ifp->if_capenable = sc->ifp->if_capabilities;
2210 sc->ifp->if_baudrate = IF_Gbps(10);
2212 #if __FreeBSD_version >= 1000000
2213 sc->ifp->if_hw_tsomax = 65536 - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN);
2214 sc->ifp->if_hw_tsomaxsegcount = OCE_MAX_TX_ELEMENTS;
2215 sc->ifp->if_hw_tsomaxsegsize = 4096;
2218 ether_ifattach(sc->ifp, sc->macaddr.mac_addr);
2225 oce_add_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2227 POCE_SOFTC sc = ifp->if_softc;
2229 if (ifp->if_softc != arg)
2231 if ((vtag == 0) || (vtag > 4095))
2234 sc->vlan_tag[vtag] = 1;
2236 if (sc->vlans_added <= (sc->max_vlans + 1))
2242 oce_del_vlan(void *arg, struct ifnet *ifp, uint16_t vtag)
2244 POCE_SOFTC sc = ifp->if_softc;
2246 if (ifp->if_softc != arg)
2248 if ((vtag == 0) || (vtag > 4095))
2251 sc->vlan_tag[vtag] = 0;
2258 * A max of 64 vlans can be configured in BE. If the user configures
2259 * more, place the card in vlan promiscuous mode.
2262 oce_vid_config(POCE_SOFTC sc)
2264 struct normal_vlan vtags[MAX_VLANFILTER_SIZE];
2265 uint16_t ntags = 0, i;
2268 if ((sc->vlans_added <= MAX_VLANFILTER_SIZE) &&
2269 (sc->ifp->if_capenable & IFCAP_VLAN_HWFILTER)) {
2270 for (i = 0; i < MAX_VLANS; i++) {
2271 if (sc->vlan_tag[i]) {
2272 vtags[ntags].vtag = i;
2277 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2278 vtags, ntags, 1, 0);
2280 status = oce_config_vlan(sc, (uint8_t) sc->if_id,
2287 oce_mac_addr_set(POCE_SOFTC sc)
2289 uint32_t old_pmac_id = sc->pmac_id;
2293 status = bcmp((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2294 sc->macaddr.size_of_struct);
2298 status = oce_mbox_macaddr_add(sc, (uint8_t *)(IF_LLADDR(sc->ifp)),
2299 sc->if_id, &sc->pmac_id);
2301 status = oce_mbox_macaddr_del(sc, sc->if_id, old_pmac_id);
2302 bcopy((IF_LLADDR(sc->ifp)), sc->macaddr.mac_addr,
2303 sc->macaddr.size_of_struct);
2306 device_printf(sc->dev, "Failed update macaddress\n");
2312 oce_handle_passthrough(struct ifnet *ifp, caddr_t data)
2314 POCE_SOFTC sc = ifp->if_softc;
2315 struct ifreq *ifr = (struct ifreq *)data;
2317 char cookie[32] = {0};
2318 void *priv_data = ifr_data_get_ptr(ifr);
2322 OCE_DMA_MEM dma_mem;
2323 struct mbx_common_get_cntl_attr *fw_cmd;
2325 if (copyin(priv_data, cookie, strlen(IOCTL_COOKIE)))
2328 if (memcmp(cookie, IOCTL_COOKIE, strlen(IOCTL_COOKIE)))
2331 ioctl_ptr = (char *)priv_data + strlen(IOCTL_COOKIE);
2332 if (copyin(ioctl_ptr, &req, sizeof(struct mbx_hdr)))
2335 req_size = le32toh(req.u0.req.request_length);
2336 if (req_size > 65536)
2339 req_size += sizeof(struct mbx_hdr);
2340 rc = oce_dma_alloc(sc, req_size, &dma_mem, 0);
2344 if (copyin(ioctl_ptr, OCE_DMAPTR(&dma_mem,char), req_size)) {
2349 rc = oce_pass_through_mbox(sc, &dma_mem, req_size);
2355 if (copyout(OCE_DMAPTR(&dma_mem,char), ioctl_ptr, req_size))
2359 firmware is filling all the attributes for this ioctl except
2360 the driver version..so fill it
2362 if(req.u0.rsp.opcode == OPCODE_COMMON_GET_CNTL_ATTRIBUTES) {
2363 fw_cmd = (struct mbx_common_get_cntl_attr *) ioctl_ptr;
2364 strncpy(fw_cmd->params.rsp.cntl_attr_info.hba_attr.drv_ver_str,
2365 COMPONENT_REVISION, strlen(COMPONENT_REVISION));
2369 oce_dma_free(sc, &dma_mem);
2375 oce_eqd_set_periodic(POCE_SOFTC sc)
2377 struct oce_set_eqd set_eqd[OCE_MAX_EQ];
2378 struct oce_aic_obj *aic;
2380 uint64_t now = 0, delta;
2381 int eqd, i, num = 0;
2382 uint32_t tx_reqs = 0, rxpkts = 0, pps;
2386 #define ticks_to_msecs(t) (1000 * (t) / hz)
2388 for (i = 0 ; i < sc->neqs; i++) {
2390 aic = &sc->aic_obj[i];
2391 /* When setting the static eq delay from the user space */
2401 rxpkts = rq->rx_stats.rx_pkts;
2404 if (i + 1 < sc->nrqs) {
2406 rxpkts += rq->rx_stats.rx_pkts;
2410 tx_reqs = wq->tx_stats.tx_reqs;
2415 if (!aic->ticks || now < aic->ticks ||
2416 rxpkts < aic->prev_rxpkts || tx_reqs < aic->prev_txreqs) {
2417 aic->prev_rxpkts = rxpkts;
2418 aic->prev_txreqs = tx_reqs;
2423 delta = ticks_to_msecs(now - aic->ticks);
2425 pps = (((uint32_t)(rxpkts - aic->prev_rxpkts) * 1000) / delta) +
2426 (((uint32_t)(tx_reqs - aic->prev_txreqs) * 1000) / delta);
2427 eqd = (pps / 15000) << 2;
2431 /* Make sure that the eq delay is in the known range */
2432 eqd = min(eqd, aic->max_eqd);
2433 eqd = max(eqd, aic->min_eqd);
2435 aic->prev_rxpkts = rxpkts;
2436 aic->prev_txreqs = tx_reqs;
2440 if (eqd != aic->cur_eqd) {
2441 set_eqd[num].delay_multiplier = (eqd * 65)/100;
2442 set_eqd[num].eq_id = eqo->eq_id;
2448 /* Is there atleast one eq that needs to be modified? */
2449 for(i = 0; i < num; i += 8) {
2451 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], 8);
2453 oce_mbox_eqd_modify_periodic(sc, &set_eqd[i], (num - i));
2458 static void oce_detect_hw_error(POCE_SOFTC sc)
2461 uint32_t ue_low = 0, ue_high = 0, ue_low_mask = 0, ue_high_mask = 0;
2462 uint32_t sliport_status = 0, sliport_err1 = 0, sliport_err2 = 0;
2469 sliport_status = OCE_READ_REG32(sc, db, SLIPORT_STATUS_OFFSET);
2470 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2471 sliport_err1 = OCE_READ_REG32(sc, db, SLIPORT_ERROR1_OFFSET);
2472 sliport_err2 = OCE_READ_REG32(sc, db, SLIPORT_ERROR2_OFFSET);
2475 ue_low = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW);
2476 ue_high = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HIGH);
2477 ue_low_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_LOW_MASK);
2478 ue_high_mask = OCE_READ_REG32(sc, devcfg, PCICFG_UE_STATUS_HI_MASK);
2480 ue_low = (ue_low & ~ue_low_mask);
2481 ue_high = (ue_high & ~ue_high_mask);
2484 /* On certain platforms BE hardware can indicate spurious UEs.
2485 * Allow the h/w to stop working completely in case of a real UE.
2486 * Hence not setting the hw_error for UE detection.
2488 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2489 sc->hw_error = TRUE;
2490 device_printf(sc->dev, "Error detected in the card\n");
2493 if (sliport_status & SLIPORT_STATUS_ERR_MASK) {
2494 device_printf(sc->dev,
2495 "ERR: sliport status 0x%x\n", sliport_status);
2496 device_printf(sc->dev,
2497 "ERR: sliport error1 0x%x\n", sliport_err1);
2498 device_printf(sc->dev,
2499 "ERR: sliport error2 0x%x\n", sliport_err2);
2503 for (i = 0; ue_low; ue_low >>= 1, i++) {
2505 device_printf(sc->dev, "UE: %s bit set\n",
2506 ue_status_low_desc[i]);
2511 for (i = 0; ue_high; ue_high >>= 1, i++) {
2513 device_printf(sc->dev, "UE: %s bit set\n",
2514 ue_status_hi_desc[i]);
2522 oce_local_timer(void *arg)
2524 POCE_SOFTC sc = arg;
2527 oce_detect_hw_error(sc);
2528 oce_refresh_nic_stats(sc);
2529 oce_refresh_queue_stats(sc);
2530 oce_mac_addr_set(sc);
2533 for (i = 0; i < sc->nwqs; i++)
2534 oce_tx_restart(sc, sc->wq[i]);
2536 /* calculate and set the eq delay for optimal interrupt rate */
2537 if (IS_BE(sc) || IS_SH(sc))
2538 oce_eqd_set_periodic(sc);
2540 callout_reset(&sc->timer, hz, oce_local_timer, sc);
2544 oce_tx_compl_clean(POCE_SOFTC sc)
2547 int i = 0, timeo = 0, num_wqes = 0;
2548 int pending_txqs = sc->nwqs;
2550 /* Stop polling for compls when HW has been silent for 10ms or
2551 * hw_error or no outstanding completions expected
2554 pending_txqs = sc->nwqs;
2556 for_all_wq_queues(sc, wq, i) {
2557 num_wqes = oce_wq_handler(wq);
2562 if(!wq->ring->num_used)
2566 if (pending_txqs == 0 || ++timeo > 10 || sc->hw_error)
2572 for_all_wq_queues(sc, wq, i) {
2573 while(wq->ring->num_used) {
2574 LOCK(&wq->tx_compl_lock);
2575 oce_process_tx_completion(wq);
2576 UNLOCK(&wq->tx_compl_lock);
2582 /* NOTE : This should only be called holding
2586 oce_if_deactivate(POCE_SOFTC sc)
2593 sc->ifp->if_drv_flags &= ~(IFF_DRV_RUNNING | IFF_DRV_OACTIVE);
2595 oce_tx_compl_clean(sc);
2597 /* Stop intrs and finish any bottom halves pending */
2598 oce_hw_intr_disable(sc);
2600 /* Since taskqueue_drain takes a Gaint Lock, We should not acquire
2601 any other lock. So unlock device lock and require after
2602 completing taskqueue_drain.
2604 UNLOCK(&sc->dev_lock);
2605 for (i = 0; i < sc->intr_count; i++) {
2606 if (sc->intrs[i].tq != NULL) {
2607 taskqueue_drain(sc->intrs[i].tq, &sc->intrs[i].task);
2610 LOCK(&sc->dev_lock);
2612 /* Delete RX queue in card with flush param */
2615 /* Invalidate any pending cq and eq entries*/
2616 for_all_evnt_queues(sc, eq, i)
2618 for_all_rq_queues(sc, rq, i)
2619 oce_drain_rq_cq(rq);
2620 for_all_wq_queues(sc, wq, i)
2621 oce_drain_wq_cq(wq);
2623 /* But still we need to get MCC aync events.
2624 So enable intrs and also arm first EQ
2626 oce_hw_intr_enable(sc);
2627 oce_arm_eq(sc, sc->eq[0]->eq_id, 0, TRUE, FALSE);
2634 oce_if_activate(POCE_SOFTC sc)
2641 sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
2643 oce_hw_intr_disable(sc);
2647 for_all_rq_queues(sc, rq, i) {
2648 rc = oce_start_rq(rq);
2650 device_printf(sc->dev, "Unable to start RX\n");
2653 for_all_wq_queues(sc, wq, i) {
2654 rc = oce_start_wq(wq);
2656 device_printf(sc->dev, "Unable to start TX\n");
2660 for_all_evnt_queues(sc, eq, i)
2661 oce_arm_eq(sc, eq->eq_id, 0, TRUE, FALSE);
2663 oce_hw_intr_enable(sc);
2668 process_link_state(POCE_SOFTC sc, struct oce_async_cqe_link_state *acqe)
2670 /* Update Link status */
2671 if ((acqe->u0.s.link_status & ~ASYNC_EVENT_LOGICAL) ==
2672 ASYNC_EVENT_LINK_UP) {
2673 sc->link_status = ASYNC_EVENT_LINK_UP;
2674 if_link_state_change(sc->ifp, LINK_STATE_UP);
2676 sc->link_status = ASYNC_EVENT_LINK_DOWN;
2677 if_link_state_change(sc->ifp, LINK_STATE_DOWN);
2682 static void oce_async_grp5_osbmc_process(POCE_SOFTC sc,
2683 struct oce_async_evt_grp5_os2bmc *evt)
2685 DW_SWAP(evt, sizeof(struct oce_async_evt_grp5_os2bmc));
2686 if (evt->u.s.mgmt_enable)
2687 sc->flags |= OCE_FLAGS_OS2BMC;
2691 sc->bmc_filt_mask = evt->u.s.arp_filter;
2692 sc->bmc_filt_mask |= (evt->u.s.dhcp_client_filt << 1);
2693 sc->bmc_filt_mask |= (evt->u.s.dhcp_server_filt << 2);
2694 sc->bmc_filt_mask |= (evt->u.s.net_bios_filt << 3);
2695 sc->bmc_filt_mask |= (evt->u.s.bcast_filt << 4);
2696 sc->bmc_filt_mask |= (evt->u.s.ipv6_nbr_filt << 5);
2697 sc->bmc_filt_mask |= (evt->u.s.ipv6_ra_filt << 6);
2698 sc->bmc_filt_mask |= (evt->u.s.ipv6_ras_filt << 7);
2699 sc->bmc_filt_mask |= (evt->u.s.mcast_filt << 8);
2703 static void oce_process_grp5_events(POCE_SOFTC sc, struct oce_mq_cqe *cqe)
2705 struct oce_async_event_grp5_pvid_state *gcqe;
2706 struct oce_async_evt_grp5_os2bmc *bmccqe;
2708 switch (cqe->u0.s.async_type) {
2709 case ASYNC_EVENT_PVID_STATE:
2711 gcqe = (struct oce_async_event_grp5_pvid_state *)cqe;
2713 sc->pvid = gcqe->tag & VLAN_VID_MASK;
2717 case ASYNC_EVENT_OS2BMC:
2718 bmccqe = (struct oce_async_evt_grp5_os2bmc *)cqe;
2719 oce_async_grp5_osbmc_process(sc, bmccqe);
2726 /* Handle the Completion Queue for the Mailbox/Async notifications */
2728 oce_mq_handler(void *arg)
2730 struct oce_mq *mq = (struct oce_mq *)arg;
2731 POCE_SOFTC sc = mq->parent;
2732 struct oce_cq *cq = mq->cq;
2733 int num_cqes = 0, evt_type = 0, optype = 0;
2734 struct oce_mq_cqe *cqe;
2735 struct oce_async_cqe_link_state *acqe;
2736 struct oce_async_event_qnq *dbgcqe;
2739 bus_dmamap_sync(cq->ring->dma.tag,
2740 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2741 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2743 while (cqe->u0.dw[3]) {
2744 DW_SWAP((uint32_t *) cqe, sizeof(oce_mq_cqe));
2745 if (cqe->u0.s.async_event) {
2746 evt_type = cqe->u0.s.event_type;
2747 optype = cqe->u0.s.async_type;
2748 if (evt_type == ASYNC_EVENT_CODE_LINK_STATE) {
2749 /* Link status evt */
2750 acqe = (struct oce_async_cqe_link_state *)cqe;
2751 process_link_state(sc, acqe);
2752 } else if (evt_type == ASYNC_EVENT_GRP5) {
2753 oce_process_grp5_events(sc, cqe);
2754 } else if (evt_type == ASYNC_EVENT_CODE_DEBUG &&
2755 optype == ASYNC_EVENT_DEBUG_QNQ) {
2756 dbgcqe = (struct oce_async_event_qnq *)cqe;
2758 sc->qnqid = dbgcqe->vlan_tag;
2759 sc->qnq_debug_event = TRUE;
2763 RING_GET(cq->ring, 1);
2764 bus_dmamap_sync(cq->ring->dma.tag,
2765 cq->ring->dma.map, BUS_DMASYNC_POSTWRITE);
2766 cqe = RING_GET_CONSUMER_ITEM_VA(cq->ring, struct oce_mq_cqe);
2771 oce_arm_cq(sc, cq->cq_id, num_cqes, FALSE);
2778 setup_max_queues_want(POCE_SOFTC sc)
2780 /* Check if it is FLEX machine. Is so dont use RSS */
2781 if ((sc->function_mode & FNM_FLEX10_MODE) ||
2782 (sc->function_mode & FNM_UMC_MODE) ||
2783 (sc->function_mode & FNM_VNIC_MODE) ||
2784 (!is_rss_enabled(sc)) ||
2789 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2790 sc->nwqs = MIN(OCE_NCPUS, sc->nrssqs);
2793 if (IS_BE2(sc) && is_rss_enabled(sc))
2794 sc->nrqs = MIN(OCE_NCPUS, sc->nrssqs) + 1;
2799 update_queues_got(POCE_SOFTC sc)
2801 if (is_rss_enabled(sc)) {
2802 sc->nrqs = sc->intr_count + 1;
2803 sc->nwqs = sc->intr_count;
2814 oce_check_ipv6_ext_hdr(struct mbuf *m)
2816 struct ether_header *eh = mtod(m, struct ether_header *);
2817 caddr_t m_datatemp = m->m_data;
2819 if (eh->ether_type == htons(ETHERTYPE_IPV6)) {
2820 m->m_data += sizeof(struct ether_header);
2821 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
2823 if((ip6->ip6_nxt != IPPROTO_TCP) && \
2824 (ip6->ip6_nxt != IPPROTO_UDP)){
2825 struct ip6_ext *ip6e = NULL;
2826 m->m_data += sizeof(struct ip6_hdr);
2828 ip6e = (struct ip6_ext *) mtod(m, struct ip6_ext *);
2829 if(ip6e->ip6e_len == 0xff) {
2830 m->m_data = m_datatemp;
2834 m->m_data = m_datatemp;
2840 is_be3_a1(POCE_SOFTC sc)
2842 if((sc->flags & OCE_FLAGS_BE3) && ((sc->asic_revision & 0xFF) < 2)) {
2848 static struct mbuf *
2849 oce_insert_vlan_tag(POCE_SOFTC sc, struct mbuf *m, boolean_t *complete)
2851 uint16_t vlan_tag = 0;
2856 /* Embed vlan tag in the packet if it is not part of it */
2857 if(m->m_flags & M_VLANTAG) {
2858 vlan_tag = EVL_VLANOFTAG(m->m_pkthdr.ether_vtag);
2859 m->m_flags &= ~M_VLANTAG;
2862 /* if UMC, ignore vlan tag insertion and instead insert pvid */
2865 vlan_tag = sc->pvid;
2871 m = ether_vlanencap(m, vlan_tag);
2875 m = ether_vlanencap(m, sc->qnqid);
2884 oce_tx_asic_stall_verify(POCE_SOFTC sc, struct mbuf *m)
2886 if(is_be3_a1(sc) && IS_QNQ_OR_UMC(sc) && \
2887 oce_check_ipv6_ext_hdr(m)) {
2894 oce_get_config(POCE_SOFTC sc)
2897 uint32_t max_rss = 0;
2899 if ((IS_BE(sc) || IS_SH(sc)) && (!sc->be3_native))
2900 max_rss = OCE_LEGACY_MODE_RSS;
2902 max_rss = OCE_MAX_RSS;
2905 rc = oce_get_profile_config(sc, max_rss);
2907 sc->nwqs = OCE_MAX_WQ;
2908 sc->nrssqs = max_rss;
2909 sc->nrqs = sc->nrssqs + 1;
2912 else { /* For BE3 don't rely on fw for determining the resources */
2913 sc->nrssqs = max_rss;
2914 sc->nrqs = sc->nrssqs + 1;
2915 sc->nwqs = OCE_MAX_WQ;
2916 sc->max_vlans = MAX_VLANFILTER_SIZE;
2921 oce_rdma_close(void)
2923 if (oce_rdma_if != NULL) {
2929 oce_get_mac_addr(POCE_SOFTC sc, uint8_t *macaddr)
2931 memcpy(macaddr, sc->macaddr.mac_addr, 6);
2935 oce_register_rdma(POCE_RDMA_INFO rdma_info, POCE_RDMA_IF rdma_if)
2938 struct oce_dev_info di;
2941 if ((rdma_info == NULL) || (rdma_if == NULL)) {
2945 if ((rdma_info->size != OCE_RDMA_INFO_SIZE) ||
2946 (rdma_if->size != OCE_RDMA_IF_SIZE)) {
2950 rdma_info->close = oce_rdma_close;
2951 rdma_info->mbox_post = oce_mbox_post;
2952 rdma_info->common_req_hdr_init = mbx_common_req_hdr_init;
2953 rdma_info->get_mac_addr = oce_get_mac_addr;
2955 oce_rdma_if = rdma_if;
2958 while (sc != NULL) {
2959 if (oce_rdma_if->announce != NULL) {
2960 memset(&di, 0, sizeof(di));
2964 di.db_bhandle = sc->db_bhandle;
2965 di.db_btag = sc->db_btag;
2966 di.db_page_size = 4096;
2967 if (sc->flags & OCE_FLAGS_USING_MSIX) {
2968 di.intr_mode = OCE_INTERRUPT_MODE_MSIX;
2969 } else if (sc->flags & OCE_FLAGS_USING_MSI) {
2970 di.intr_mode = OCE_INTERRUPT_MODE_MSI;
2972 di.intr_mode = OCE_INTERRUPT_MODE_INTX;
2974 di.dev_family = OCE_GEN2_FAMILY; // fixme: must detect skyhawk
2975 if (di.intr_mode != OCE_INTERRUPT_MODE_INTX) {
2976 di.msix.num_vectors = sc->intr_count + sc->roce_intr_count;
2977 di.msix.start_vector = sc->intr_count;
2978 for (i=0; i<di.msix.num_vectors; i++) {
2979 di.msix.vector_list[i] = sc->intrs[i].vector;
2983 memcpy(di.mac_addr, sc->macaddr.mac_addr, 6);
2984 di.vendor_id = pci_get_vendor(sc->dev);
2985 di.dev_id = pci_get_device(sc->dev);
2987 if (sc->rdma_flags & OCE_RDMA_FLAG_SUPPORTED) {
2988 di.flags |= OCE_RDMA_INFO_RDMA_SUPPORTED;
2991 rdma_if->announce(&di);
3000 oce_read_env_variables( POCE_SOFTC sc )
3005 /* read if user wants to enable hwlro or swlro */
3006 //value = getenv("oce_enable_hwlro");
3007 if(value && IS_SH(sc)) {
3008 sc->enable_hwlro = strtol(value, NULL, 10);
3009 if(sc->enable_hwlro) {
3010 rc = oce_mbox_nic_query_lro_capabilities(sc, NULL, NULL);
3012 device_printf(sc->dev, "no hardware lro support\n");
3013 device_printf(sc->dev, "software lro enabled\n");
3014 sc->enable_hwlro = 0;
3016 device_printf(sc->dev, "hardware lro enabled\n");
3017 oce_max_rsp_handled = 32;
3020 device_printf(sc->dev, "software lro enabled\n");
3023 sc->enable_hwlro = 0;
3026 /* read mbuf size */
3027 //value = getenv("oce_rq_buf_size");
3028 if(value && IS_SH(sc)) {
3029 oce_rq_buf_size = strtol(value, NULL, 10);
3030 switch(oce_rq_buf_size) {
3038 device_printf(sc->dev, " Supported oce_rq_buf_size values are 2K, 4K, 9K, 16K \n");
3039 oce_rq_buf_size = 2048;