Date: Thu, 13 May 2010 17:45:32 +0000 (UTC) From: Navdeep Parhar <np@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-8@freebsd.org Subject: svn commit: r208039 - stable/8/sys/dev/cxgb Message-ID: <201005131745.o4DHjW7b004078@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: np Date: Thu May 13 17:45:32 2010 New Revision: 208039 URL: http://svn.freebsd.org/changeset/base/208039 Log: MFC r207639, r207643, r207687, r207688 r207639: Add IFCAP_LINKSTATE to cxgb r207643: T3 hardware filters + cxgbtool interface to this functionality. r207687: if_cxgb should not depend on the T3 firmware module directly. r207688: various doorbell adjustments Modified: stable/8/sys/dev/cxgb/cxgb_adapter.h stable/8/sys/dev/cxgb/cxgb_ioctl.h stable/8/sys/dev/cxgb/cxgb_main.c stable/8/sys/dev/cxgb/cxgb_sge.c Directory Properties: stable/8/sys/ (props changed) stable/8/sys/amd64/include/xen/ (props changed) stable/8/sys/cddl/contrib/opensolaris/ (props changed) stable/8/sys/contrib/dev/acpica/ (props changed) stable/8/sys/contrib/pf/ (props changed) stable/8/sys/dev/xen/xenpci/ (props changed) stable/8/sys/geom/sched/ (props changed) Modified: stable/8/sys/dev/cxgb/cxgb_adapter.h ============================================================================== --- stable/8/sys/dev/cxgb/cxgb_adapter.h Thu May 13 17:32:30 2010 (r208038) +++ stable/8/sys/dev/cxgb/cxgb_adapter.h Thu May 13 17:45:32 2010 (r208039) @@ -204,6 +204,7 @@ struct sge_fl { uint32_t cidx; uint32_t pidx; uint32_t gen; + uint32_t db_pending; bus_addr_t phys_addr; uint32_t cntxt_id; uint32_t empty; @@ -232,6 +233,7 @@ struct sge_txq { uint32_t pidx; uint32_t gen; uint32_t unacked; + uint32_t db_pending; struct tx_desc *desc; struct tx_sw_desc *sdesc; uint32_t token; Modified: stable/8/sys/dev/cxgb/cxgb_ioctl.h ============================================================================== --- stable/8/sys/dev/cxgb/cxgb_ioctl.h Thu May 13 17:32:30 2010 (r208038) +++ stable/8/sys/dev/cxgb/cxgb_ioctl.h Thu May 13 17:45:32 2010 (r208039) @@ -59,6 +59,9 @@ enum { CH_CLEAR_STATS, CH_GET_UP_LA, CH_GET_UP_IOQS, + CH_SET_FILTER, + CH_DEL_FILTER, + CH_GET_FILTER, }; /* statistics categories */ @@ -215,6 +218,29 @@ struct ch_up_ioqs { struct t3_ioq_entry *data; }; +struct ch_filter_tuple { + uint32_t sip; + uint32_t dip; + uint16_t sport; + uint16_t dport; + uint16_t vlan:12; + uint16_t vlan_prio:3; +}; + +struct ch_filter { + uint32_t filter_id; + struct ch_filter_tuple val; + struct ch_filter_tuple mask; + uint16_t mac_addr_idx; + uint8_t mac_hit:1; + uint8_t proto:2; + + uint8_t want_filter_id:1; + uint8_t pass:1; + uint8_t rss:1; + uint8_t qset; +}; + #define CHELSIO_SETREG _IOW('f', CH_SETREG, struct ch_reg) #define CHELSIO_GETREG _IOWR('f', CH_GETREG, struct ch_reg) #define CHELSIO_GETMTUTAB _IOR('f', CH_GETMTUTAB, struct ch_mtus) @@ -239,4 +265,7 @@ struct ch_up_ioqs { #define CHELSIO_GET_EEPROM _IOWR('f', CH_GET_EEPROM, struct ch_eeprom) #define CHELSIO_GET_UP_LA _IOWR('f', CH_GET_UP_LA, struct ch_up_la) #define CHELSIO_GET_UP_IOQS _IOWR('f', CH_GET_UP_IOQS, struct ch_up_ioqs) +#define CHELSIO_SET_FILTER _IOW('f', CH_SET_FILTER, struct ch_filter) +#define CHELSIO_DEL_FILTER _IOW('f', CH_DEL_FILTER, struct ch_filter) +#define CHELSIO_GET_FILTER _IOWR('f', CH_GET_FILTER, struct ch_filter) #endif Modified: stable/8/sys/dev/cxgb/cxgb_main.c ============================================================================== --- stable/8/sys/dev/cxgb/cxgb_main.c Thu May 13 17:32:30 2010 (r208038) +++ stable/8/sys/dev/cxgb/cxgb_main.c Thu May 13 17:45:32 2010 (r208039) @@ -99,6 +99,13 @@ static void cxgb_ext_intr_handler(void * static void cxgb_tick_handler(void *, int); static void cxgb_tick(void *); static void setup_rss(adapter_t *sc); +static int alloc_filters(struct adapter *); +static int setup_hw_filters(struct adapter *); +static int set_filter(struct adapter *, int, const struct filter_info *); +static inline void mk_set_tcb_field(struct cpl_set_tcb_field *, unsigned int, + unsigned int, u64, u64); +static inline void set_tcb_field_ulp(struct cpl_set_tcb_field *, unsigned int, + unsigned int, u64, u64); /* Attachment glue for the PCI controller end of the device. Each port of * the device is attached separately, as defined later. @@ -981,7 +988,7 @@ cxgb_makedev(struct port_info *pi) #define CXGB_CAP (IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | \ IFCAP_VLAN_HWCSUM | IFCAP_TSO | IFCAP_JUMBO_MTU | IFCAP_LRO | \ - IFCAP_VLAN_HWTSO) + IFCAP_VLAN_HWTSO | IFCAP_LINKSTATE) #define CXGB_CAP_ENABLE (CXGB_CAP & ~IFCAP_TSO6) static int @@ -1665,6 +1672,13 @@ cxgb_up(struct adapter *sc) if ((err = update_tpsram(sc))) goto out; + if (is_offload(sc)) { + sc->params.mc5.nservers = 0; + sc->params.mc5.nroutes = 0; + sc->params.mc5.nfilters = t3_mc5_size(&sc->mc5) - + MC5_MIN_TIDS; + } + err = t3_init_hw(sc, 0); if (err) goto out; @@ -1676,6 +1690,7 @@ cxgb_up(struct adapter *sc) if (err) goto out; + alloc_filters(sc); setup_rss(sc); t3_intr_clear(sc); @@ -1702,6 +1717,7 @@ cxgb_up(struct adapter *sc) if (!(sc->flags & QUEUES_BOUND)) { bind_qsets(sc); + setup_hw_filters(sc); sc->flags |= QUEUES_BOUND; } @@ -3080,6 +3096,139 @@ cxgb_extension_ioctl(struct cdev *dev, u free(buf, M_DEVBUF); break; } + case CHELSIO_SET_FILTER: { + struct ch_filter *f = (struct ch_filter *)data;; + struct filter_info *p; + unsigned int nfilters = sc->params.mc5.nfilters; + + if (!is_offload(sc)) + return (EOPNOTSUPP); /* No TCAM */ + if (!(sc->flags & FULL_INIT_DONE)) + return (EAGAIN); /* mc5 not setup yet */ + if (nfilters == 0) + return (EBUSY); /* TOE will use TCAM */ + + /* sanity checks */ + if (f->filter_id >= nfilters || + (f->val.dip && f->mask.dip != 0xffffffff) || + (f->val.sport && f->mask.sport != 0xffff) || + (f->val.dport && f->mask.dport != 0xffff) || + (f->val.vlan && f->mask.vlan != 0xfff) || + (f->val.vlan_prio && + f->mask.vlan_prio != FILTER_NO_VLAN_PRI) || + (f->mac_addr_idx != 0xffff && f->mac_addr_idx > 15) || + f->qset >= SGE_QSETS || + sc->rrss_map[f->qset] >= RSS_TABLE_SIZE) + return (EINVAL); + + /* Was allocated with M_WAITOK */ + KASSERT(sc->filters, ("filter table NULL\n")); + + p = &sc->filters[f->filter_id]; + if (p->locked) + return (EPERM); + + bzero(p, sizeof(*p)); + p->sip = f->val.sip; + p->sip_mask = f->mask.sip; + p->dip = f->val.dip; + p->sport = f->val.sport; + p->dport = f->val.dport; + p->vlan = f->mask.vlan ? f->val.vlan : 0xfff; + p->vlan_prio = f->mask.vlan_prio ? (f->val.vlan_prio & 6) : + FILTER_NO_VLAN_PRI; + p->mac_hit = f->mac_hit; + p->mac_vld = f->mac_addr_idx != 0xffff; + p->mac_idx = f->mac_addr_idx; + p->pkt_type = f->proto; + p->report_filter_id = f->want_filter_id; + p->pass = f->pass; + p->rss = f->rss; + p->qset = f->qset; + + error = set_filter(sc, f->filter_id, p); + if (error == 0) + p->valid = 1; + break; + } + case CHELSIO_DEL_FILTER: { + struct ch_filter *f = (struct ch_filter *)data; + struct filter_info *p; + unsigned int nfilters = sc->params.mc5.nfilters; + + if (!is_offload(sc)) + return (EOPNOTSUPP); + if (!(sc->flags & FULL_INIT_DONE)) + return (EAGAIN); + if (nfilters == 0 || sc->filters == NULL) + return (EINVAL); + if (f->filter_id >= nfilters) + return (EINVAL); + + p = &sc->filters[f->filter_id]; + if (p->locked) + return (EPERM); + if (!p->valid) + return (EFAULT); /* Read "Bad address" as "Bad index" */ + + bzero(p, sizeof(*p)); + p->sip = p->sip_mask = 0xffffffff; + p->vlan = 0xfff; + p->vlan_prio = FILTER_NO_VLAN_PRI; + p->pkt_type = 1; + error = set_filter(sc, f->filter_id, p); + break; + } + case CHELSIO_GET_FILTER: { + struct ch_filter *f = (struct ch_filter *)data; + struct filter_info *p; + unsigned int i, nfilters = sc->params.mc5.nfilters; + + if (!is_offload(sc)) + return (EOPNOTSUPP); + if (!(sc->flags & FULL_INIT_DONE)) + return (EAGAIN); + if (nfilters == 0 || sc->filters == NULL) + return (EINVAL); + + i = f->filter_id == 0xffffffff ? 0 : f->filter_id + 1; + for (; i < nfilters; i++) { + p = &sc->filters[i]; + if (!p->valid) + continue; + + bzero(f, sizeof(*f)); + + f->filter_id = i; + f->val.sip = p->sip; + f->mask.sip = p->sip_mask; + f->val.dip = p->dip; + f->mask.dip = p->dip ? 0xffffffff : 0; + f->val.sport = p->sport; + f->mask.sport = p->sport ? 0xffff : 0; + f->val.dport = p->dport; + f->mask.dport = p->dport ? 0xffff : 0; + f->val.vlan = p->vlan == 0xfff ? 0 : p->vlan; + f->mask.vlan = p->vlan == 0xfff ? 0 : 0xfff; + f->val.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ? + 0 : p->vlan_prio; + f->mask.vlan_prio = p->vlan_prio == FILTER_NO_VLAN_PRI ? + 0 : FILTER_NO_VLAN_PRI; + f->mac_hit = p->mac_hit; + f->mac_addr_idx = p->mac_vld ? p->mac_idx : 0xffff; + f->proto = p->pkt_type; + f->want_filter_id = p->report_filter_id; + f->pass = p->pass; + f->rss = p->rss; + f->qset = p->qset; + + break; + } + + if (i == nfilters) + f->filter_id = 0xffffffff; + break; + } default: return (EOPNOTSUPP); break; @@ -3134,5 +3283,127 @@ cxgb_get_regs(adapter_t *sc, struct ch_i XGM_REG(A_XGM_RX_SPI4_SOP_EOP_CNT, 1)); } +static int +alloc_filters(struct adapter *sc) +{ + struct filter_info *p; + unsigned int nfilters = sc->params.mc5.nfilters; + + if (nfilters == 0) + return (0); + + p = malloc(sizeof(*p) * nfilters, M_DEVBUF, M_WAITOK | M_ZERO); + sc->filters = p; + + p = &sc->filters[nfilters - 1]; + p->vlan = 0xfff; + p->vlan_prio = FILTER_NO_VLAN_PRI; + p->pass = p->rss = p->valid = p->locked = 1; + + return (0); +} + +static int +setup_hw_filters(struct adapter *sc) +{ + int i, rc; + unsigned int nfilters = sc->params.mc5.nfilters; + + if (!sc->filters) + return (0); + + t3_enable_filters(sc); + + for (i = rc = 0; i < nfilters && !rc; i++) { + if (sc->filters[i].locked) + rc = set_filter(sc, i, &sc->filters[i]); + } + + return (rc); +} + +static int +set_filter(struct adapter *sc, int id, const struct filter_info *f) +{ + int len; + struct mbuf *m; + struct ulp_txpkt *txpkt; + struct work_request_hdr *wr; + struct cpl_pass_open_req *oreq; + struct cpl_set_tcb_field *sreq; + + len = sizeof(*wr) + sizeof(*oreq) + 2 * sizeof(*sreq); + KASSERT(len <= MHLEN, ("filter request too big for an mbuf")); + + id += t3_mc5_size(&sc->mc5) - sc->params.mc5.nroutes - + sc->params.mc5.nfilters; + + m = m_gethdr(M_WAITOK, MT_DATA); + m->m_len = m->m_pkthdr.len = len; + bzero(mtod(m, char *), len); -MODULE_DEPEND(if_cxgb, cxgb_t3fw, 1, 1, 1); + wr = mtod(m, struct work_request_hdr *); + wr->wrh_hi = htonl(V_WR_OP(FW_WROPCODE_BYPASS) | F_WR_ATOMIC); + + oreq = (struct cpl_pass_open_req *)(wr + 1); + txpkt = (struct ulp_txpkt *)oreq; + txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT)); + txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*oreq) / 8)); + OPCODE_TID(oreq) = htonl(MK_OPCODE_TID(CPL_PASS_OPEN_REQ, id)); + oreq->local_port = htons(f->dport); + oreq->peer_port = htons(f->sport); + oreq->local_ip = htonl(f->dip); + oreq->peer_ip = htonl(f->sip); + oreq->peer_netmask = htonl(f->sip_mask); + oreq->opt0h = 0; + oreq->opt0l = htonl(F_NO_OFFLOAD); + oreq->opt1 = htonl(V_MAC_MATCH_VALID(f->mac_vld) | + V_CONN_POLICY(CPL_CONN_POLICY_FILTER) | + V_VLAN_PRI(f->vlan_prio >> 1) | + V_VLAN_PRI_VALID(f->vlan_prio != FILTER_NO_VLAN_PRI) | + V_PKT_TYPE(f->pkt_type) | V_OPT1_VLAN(f->vlan) | + V_MAC_MATCH(f->mac_idx | (f->mac_hit << 4))); + + sreq = (struct cpl_set_tcb_field *)(oreq + 1); + set_tcb_field_ulp(sreq, id, 1, 0x1800808000ULL, + (f->report_filter_id << 15) | (1 << 23) | + ((u64)f->pass << 35) | ((u64)!f->rss << 36)); + set_tcb_field_ulp(sreq + 1, id, 0, 0xffffffff, (2 << 19) | 1); + t3_mgmt_tx(sc, m); + + if (f->pass && !f->rss) { + len = sizeof(*sreq); + m = m_gethdr(M_WAITOK, MT_DATA); + m->m_len = m->m_pkthdr.len = len; + bzero(mtod(m, char *), len); + sreq = mtod(m, struct cpl_set_tcb_field *); + sreq->wr.wrh_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD)); + mk_set_tcb_field(sreq, id, 25, 0x3f80000, + (u64)sc->rrss_map[f->qset] << 19); + t3_mgmt_tx(sc, m); + } + return 0; +} + +static inline void +mk_set_tcb_field(struct cpl_set_tcb_field *req, unsigned int tid, + unsigned int word, u64 mask, u64 val) +{ + OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_SET_TCB_FIELD, tid)); + req->reply = V_NO_REPLY(1); + req->cpu_idx = 0; + req->word = htons(word); + req->mask = htobe64(mask); + req->val = htobe64(val); +} + +static inline void +set_tcb_field_ulp(struct cpl_set_tcb_field *req, unsigned int tid, + unsigned int word, u64 mask, u64 val) +{ + struct ulp_txpkt *txpkt = (struct ulp_txpkt *)req; + + txpkt->cmd_dest = htonl(V_ULPTX_CMD(ULP_TXPKT)); + txpkt->len = htonl(V_ULPTX_NFLITS(sizeof(*req) / 8)); + mk_set_tcb_field(req, tid, word, mask, val); +} Modified: stable/8/sys/dev/cxgb/cxgb_sge.c ============================================================================== --- stable/8/sys/dev/cxgb/cxgb_sge.c Thu May 13 17:32:30 2010 (r208038) +++ stable/8/sys/dev/cxgb/cxgb_sge.c Thu May 13 17:45:32 2010 (r208039) @@ -696,7 +696,7 @@ refill_fl(adapter_t *sc, struct sge_fl * struct refill_fl_cb_arg cb_arg; struct mbuf *m; caddr_t cl; - int err, count = 0; + int err; cb_arg.error = 0; while (n--) { @@ -754,12 +754,14 @@ refill_fl(adapter_t *sc, struct sge_fl * d = q->desc; } q->credits++; - count++; + q->db_pending++; } done: - if (count) + if (q->db_pending >= 32) { + q->db_pending = 0; t3_write_reg(sc, A_SG_KDOORBELL, V_EGRCNTX(q->cntxt_id)); + } } @@ -810,8 +812,10 @@ __refill_fl(adapter_t *adap, struct sge_ static __inline void __refill_fl_lt(adapter_t *adap, struct sge_fl *fl, int max) { - if ((fl->size - fl->credits) < max) - refill_fl(adap, fl, min(max, fl->size - fl->credits)); + uint32_t reclaimable = fl->size - fl->credits; + + if (reclaimable > 0) + refill_fl(adap, fl, min(max, reclaimable)); } /** @@ -1261,7 +1265,7 @@ make_sgl(struct sg_ent *sgp, bus_dma_seg * When GTS is disabled we unconditionally ring the doorbell. */ static __inline void -check_ring_tx_db(adapter_t *adap, struct sge_txq *q) +check_ring_tx_db(adapter_t *adap, struct sge_txq *q, int mustring) { #if USE_GTS clear_bit(TXQ_LAST_PKT_DB, &q->flags); @@ -1275,9 +1279,12 @@ check_ring_tx_db(adapter_t *adap, struct F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id)); } #else - wmb(); /* write descriptors before telling HW */ - t3_write_reg(adap, A_SG_KDOORBELL, - F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id)); + if (mustring || ++q->db_pending >= 32) { + wmb(); /* write descriptors before telling HW */ + t3_write_reg(adap, A_SG_KDOORBELL, + F_SELEGRCNTX | V_EGRCNTX(q->cntxt_id)); + q->db_pending = 0; + } #endif } @@ -1480,7 +1487,7 @@ t3_encap(struct sge_qset *qs, struct mbu wmb(); ETHER_BPF_MTAP(pi->ifp, m0); wr_gen2(txd, txqs.gen); - check_ring_tx_db(sc, txq); + check_ring_tx_db(sc, txq, 0); return (0); } else if (tso_info) { int eth_type; @@ -1543,7 +1550,7 @@ t3_encap(struct sge_qset *qs, struct mbu wmb(); ETHER_BPF_MTAP(pi->ifp, m0); wr_gen2(txd, txqs.gen); - check_ring_tx_db(sc, txq); + check_ring_tx_db(sc, txq, 0); m_freem(m0); return (0); } @@ -1574,7 +1581,7 @@ t3_encap(struct sge_qset *qs, struct mbu wmb(); ETHER_BPF_MTAP(pi->ifp, m0); wr_gen2(txd, txqs.gen); - check_ring_tx_db(sc, txq); + check_ring_tx_db(sc, txq, 0); m_freem(m0); return (0); } @@ -1593,7 +1600,7 @@ t3_encap(struct sge_qset *qs, struct mbu wr_lo = htonl(V_WR_TID(txq->token)); write_wr_hdr_sgl(ndesc, txd, &txqs, txq, sgl, flits, sgl_flits, wr_hi, wr_lo); - check_ring_tx_db(sc, txq); + check_ring_tx_db(sc, txq, 0); return (0); } @@ -1643,7 +1650,6 @@ cxgb_start_locked(struct sge_qset *qs) { struct mbuf *m_head = NULL; struct sge_txq *txq = &qs->txq[TXQ_ETH]; - int in_use_init = txq->in_use; struct port_info *pi = qs->port; struct ifnet *ifp = pi->ifp; @@ -1655,8 +1661,7 @@ cxgb_start_locked(struct sge_qset *qs) return; } TXQ_LOCK_ASSERT(qs); - while ((txq->in_use - in_use_init < TX_START_MAX_DESC) && - !TXQ_RING_EMPTY(qs) && (ifp->if_drv_flags & IFF_DRV_RUNNING) && + while (!TXQ_RING_EMPTY(qs) && (ifp->if_drv_flags & IFF_DRV_RUNNING) && pi->link_config.link_ok) { reclaim_completed_tx(qs, cxgb_tx_reclaim_threshold, TXQ_ETH); @@ -1674,6 +1679,10 @@ cxgb_start_locked(struct sge_qset *qs) m_head = NULL; } + + if (txq->db_pending) + check_ring_tx_db(pi->adapter, txq, 1); + if (!TXQ_RING_EMPTY(qs) && callout_pending(&txq->txq_timer) == 0 && pi->link_config.link_ok) callout_reset_on(&txq->txq_timer, 1, cxgb_tx_timeout, @@ -1707,6 +1716,9 @@ cxgb_transmit_locked(struct ifnet *ifp, (error = drbr_enqueue(ifp, br, m)) != 0) return (error); } else { + if (txq->db_pending) + check_ring_tx_db(pi->adapter, txq, 1); + /* * We've bypassed the buf ring so we need to update * the stats directly @@ -2354,7 +2366,7 @@ again: reclaim_completed_tx(qs, 16, TXQ_ TXQ_UNLOCK(qs); write_ofld_wr(adap, m, q, pidx, gen, ndesc, segs, nsegs); - check_ring_tx_db(adap, q); + check_ring_tx_db(adap, q, 1); return (0); } @@ -3033,7 +3045,7 @@ process_responses(adapter_t *adap, struc r = rspq->desc; } - if (++rspq->credits >= (rspq->size / 4)) { + if (++rspq->credits >= 64) { refill_rspq(adap, rspq, rspq->credits); rspq->credits = 0; }
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201005131745.o4DHjW7b004078>