Date: Thu, 12 May 2011 23:26:53 +0000 (UTC) From: David Christensen <davidch@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-head@freebsd.org Subject: svn commit: r221826 - head/sys/dev/bxe Message-ID: <201105122326.p4CNQrXa053532@svn.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: davidch Date: Thu May 12 23:26:53 2011 New Revision: 221826 URL: http://svn.freebsd.org/changeset/base/221826 Log: - Use bus_describe_intr() to describe interrupt usage. - Use bus_bind_intr() to bind interrupt to a CPU when RSS/TSS is used. - Use M_DONTWAIT for RSS/TSS buffer allocation. - Add statistic to track max DRBR queue depth. - Fix problem in bxe_change_mtu() which referenced the old MTU size in a debug print statement. MFC after: Two weeks Modified: head/sys/dev/bxe/if_bxe.c head/sys/dev/bxe/if_bxe.h Modified: head/sys/dev/bxe/if_bxe.c ============================================================================== --- head/sys/dev/bxe/if_bxe.c Thu May 12 22:31:13 2011 (r221825) +++ head/sys/dev/bxe/if_bxe.c Thu May 12 23:26:53 2011 (r221826) @@ -1330,9 +1330,9 @@ bxe_interrupt_attach(struct bxe_softc *s /* Setup the slowpath deferred task queue. */ TASK_INIT(&sc->task, 0, bxe_task_sp, sc); sc->tq = taskqueue_create_fast("bxe_spq", M_NOWAIT, - taskqueue_thread_enqueue, &sc->tq); + taskqueue_thread_enqueue, &sc->tq); taskqueue_start_threads(&sc->tq, 1, PI_NET, "%s spq", - device_get_nameunit(sc->dev)); + device_get_nameunit(sc->dev)); #endif /* Setup interrupt handlers. */ @@ -1359,13 +1359,19 @@ bxe_interrupt_attach(struct bxe_softc *s goto bxe_interrupt_attach_exit; } +#if __FreeBSD_version >= 800504 + bus_describe_intr(sc->dev, + sc->bxe_msix_res[0], + sc->bxe_msix_tag[0], + "sp"); +#endif + /* Now initialize the fastpath vectors. */ for (i = 0; i < (sc->num_queues); i++) { fp = &sc->fp[i]; - DBPRINT(sc, - (BXE_VERBOSE_LOAD | BXE_VERBOSE_INTR), - "%s(): Enabling MSI-X[%d] vector.\n", - __FUNCTION__, i + 1); + DBPRINT(sc, (BXE_VERBOSE_LOAD | BXE_VERBOSE_INTR), + "%s(): Enabling MSI-X[%d] vector.\n", + __FUNCTION__, i + 1); /* * Setup the interrupt handler. Note that we pass the * fastpath context to the interrupt handler in this @@ -1377,8 +1383,7 @@ bxe_interrupt_attach(struct bxe_softc *s NULL, bxe_intr_fp, fp, - &sc->bxe_msix_tag[i + 1] - ); + &sc->bxe_msix_tag[i + 1]); if (rc) { BXE_PRINTF( @@ -1386,6 +1391,21 @@ bxe_interrupt_attach(struct bxe_softc *s __FILE__, __LINE__, (i + 1)); goto bxe_interrupt_attach_exit; } + +#if __FreeBSD_version >= 800504 + bus_describe_intr(sc->dev, + sc->bxe_msix_res[i + 1], + sc->bxe_msix_tag[i + 1], + "fp[%02d]", + i); +#endif + + /* Bind the fastpath instance to a CPU. */ + if (sc->num_queues > 1) { + bus_bind_intr(sc->dev, + sc->bxe_msix_res[i + 1], i); + } + #ifdef BXE_TASK TASK_INIT(&fp->task, 0, bxe_task_fp, fp); fp->tq = taskqueue_create_fast("bxe_fpq", M_NOWAIT, @@ -1418,6 +1438,13 @@ bxe_interrupt_attach(struct bxe_softc *s goto bxe_interrupt_attach_exit; } +#if __FreeBSD_version >= 800504 + bus_describe_intr(sc->dev, + sc->bxe_msi_res[0], + sc->bxe_msi_tag[0], + "sp"); +#endif + /* Now initialize the fastpath vectors. */ for (i = 0; i < (sc->num_queues); i++) { fp = &sc->fp[i]; @@ -1445,6 +1472,15 @@ bxe_interrupt_attach(struct bxe_softc *s __FILE__, __LINE__, (i + 1)); goto bxe_interrupt_attach_exit; } + +#if __FreeBSD_version >= 800504 + bus_describe_intr(sc->dev, + sc->bxe_msi_res[i + 1], + sc->bxe_msi_tag[i + 1], + "fp[%02d]", + i); +#endif + #ifdef BXE_TASK TASK_INIT(&fp->task, 0, bxe_task_fp, fp); fp->tq = taskqueue_create_fast("bxe_fpq", M_NOWAIT, @@ -3646,7 +3682,7 @@ bxe_alloc_buf_rings(struct bxe_softc *sc if (fp != NULL) { fp->br = buf_ring_alloc(BXE_BR_SIZE, - M_DEVBUF, M_WAITOK, &fp->mtx); + M_DEVBUF, M_DONTWAIT, &fp->mtx); if (fp->br == NULL) { rc = ENOMEM; return(rc); @@ -9404,7 +9440,7 @@ bxe_tx_mq_start_exit: /* - * Multiqueue (RSS) transmit routine. + * Multiqueue (TSS) transmit routine. * * Returns: * 0 if transmit succeeds, !0 otherwise. @@ -9415,14 +9451,18 @@ bxe_tx_mq_start_locked(struct ifnet *ifp { struct bxe_softc *sc; struct mbuf *next; - int rc = 0, tx_count = 0; + int depth, rc = 0, tx_count = 0; sc = fp->sc; DBENTER(BXE_EXTREME_SEND); + depth = drbr_inuse(ifp, fp->br); + if (depth > fp->max_drbr_queue_depth) { + fp->max_drbr_queue_depth = depth; + } DBPRINT(sc, BXE_EXTREME_SEND, "%s(): fp[%02d], drbr queue depth=%d\n", - __FUNCTION__, fp->index, drbr_inuse(ifp, fp->br)); + __FUNCTION__, fp->index, depth); BXE_FP_LOCK_ASSERT(fp); @@ -10509,11 +10549,11 @@ bxe_alloc_mbuf(struct bxe_fastpath *fp, /* Check whether the allocation succeeded and handle a failure. */ if (__predict_false(m_new == NULL)) { - DBPRINT(sc, BXE_WARN, "%s(): mbuf allocation failure!\n", - __FUNCTION__); + DBPRINT(sc, BXE_WARN, "%s(): Failed to allocate %d byte " + "mbuf on fp[%02d]!\n", __FUNCTION__, size, fp->index); fp->mbuf_alloc_failed++; - goto bxe_alloc_mbuf_exit; - } + goto bxe_alloc_mbuf_exit; + } /* Do a little extra error checking when debugging. */ DBRUN(M_ASSERTPKTHDR(m_new)); @@ -10556,7 +10596,7 @@ bxe_map_mbuf(struct bxe_fastpath *fp, st __FUNCTION__); sc->debug_mbuf_sim_map_failed++; fp->mbuf_alloc_failed++; - DBRUN(sc->debug_memory_allocated -= m->m_len); + sc->debug_memory_allocated -= m->m_len; m_freem(m); rc = EINVAL; goto bxe_map_mbuf_exit; @@ -10568,10 +10608,11 @@ bxe_map_mbuf(struct bxe_fastpath *fp, st /* Handle any mapping errors. */ if (__predict_false(rc)) { - DBPRINT(sc, BXE_WARN, "%s(): mbuf mapping failure (%d)!\n", - __FUNCTION__, rc); - m_freem(m); + DBPRINT(sc, BXE_WARN, "%s(): mbuf mapping failure (%d) on " + "fp[%02d]!\n", __FUNCTION__, rc, fp->index); fp->mbuf_alloc_failed++; + DBRUN(sc->debug_memory_allocated -= m->m_len); + m_freem(m); goto bxe_map_mbuf_exit; } @@ -10583,6 +10624,7 @@ bxe_map_mbuf(struct bxe_fastpath *fp, st *seg = segs[0]; bxe_map_mbuf_exit: + DBEXIT(BXE_INSANE); return (rc); } @@ -10961,6 +11003,19 @@ bxe_init_rx_chains(struct bxe_softc *sc) } } + /* + * ToDo: Need a cleanup path if memory allocation + * fails during initializtion. This is especially + * easy if multiqueue is used on a system with + * jumbo frames and many CPUs. On my 16GB system + * with 8 CPUs I get the following defaults: + * + * kern.ipc.nmbjumbo16: 3200 + * kern.ipc.nmbjumbo9: 6400 + * kern.ipc.nmbjumbop: 12800 + * kern.ipc.nmbclusters: 25600 + */ + DBEXIT(BXE_VERBOSE_LOAD | BXE_VERBOSE_RESET); } @@ -15554,7 +15609,7 @@ bxe_txeof(struct bxe_fastpath *fp) #endif txbd = - &fp->tx_bd_chain[TX_PAGE(sw_tx_chain_cons)][TX_IDX(sw_tx_chain_cons)].start_bd; +&fp->tx_bd_chain[TX_PAGE(sw_tx_chain_cons)][TX_IDX(sw_tx_chain_cons)].start_bd; #ifdef BXE_DEBUG if (txbd == NULL) { @@ -15764,7 +15819,7 @@ bxe_change_mtu(struct bxe_softc *sc, int sc->bxe_ifp->if_mtu = ifp->if_mtu; if (if_drv_running) { DBPRINT(sc, BXE_INFO_IOCTL, "%s(): Changing the MTU to %d.\n", - __FUNCTION__, sc->port.ether_mtu); + __FUNCTION__, sc->bxe_ifp->if_mtu); bxe_stop_locked(sc, UNLOAD_NORMAL); bxe_init_locked(sc, LOAD_NORMAL); @@ -16315,6 +16370,12 @@ bxe_add_sysctls(struct bxe_softc *sc) "tx_chain_lost_mbuf", CTLFLAG_RD, &fp->tx_chain_lost_mbuf, "Mbufs lost on TX chain count"); + + SYSCTL_ADD_INT(ctx, queue_list, OID_AUTO, + "max_drbr_queue_depth", + CTLFLAG_RD, &fp->max_drbr_queue_depth, + 0, "Driver queue maximum dpeth"); + #ifdef BXE_DEBUG SYSCTL_ADD_ULONG(ctx, queue_list, OID_AUTO, "null_cqe_flags", Modified: head/sys/dev/bxe/if_bxe.h ============================================================================== --- head/sys/dev/bxe/if_bxe.h Thu May 12 22:31:13 2011 (r221825) +++ head/sys/dev/bxe/if_bxe.h Thu May 12 23:26:53 2011 (r221826) @@ -1110,6 +1110,8 @@ struct bxe_fastpath { int sge_mbuf_alloc; int tpa_mbuf_alloc; + int max_drbr_queue_depth; + uint64_t tpa_queue_used; unsigned long null_cqe_flags; @@ -1148,7 +1150,6 @@ struct bxe_fastpath { #define BXE_TX_CHAIN_PAGE_SZ BCM_PAGE_SIZE #define BXE_RX_CHAIN_PAGE_SZ BCM_PAGE_SIZE -/* ToDo: Audit this structure for unused varaibles. */ struct bxe_softc { struct ifnet *bxe_ifp; int media;
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201105122326.p4CNQrXa053532>