Date: Mon, 13 Jun 2016 07:30:55 +0000 (UTC) From: Sepherosa Ziehau <sephe@FreeBSD.org> To: src-committers@freebsd.org, svn-src-all@freebsd.org, svn-src-stable@freebsd.org, svn-src-stable-10@freebsd.org Subject: svn commit: r301861 - stable/10/sys/dev/hyperv/netvsc Message-ID: <201606130730.u5D7Utc6086560@repo.freebsd.org>
next in thread | raw e-mail | index | archive | help
Author: sephe Date: Mon Jun 13 07:30:54 2016 New Revision: 301861 URL: https://svnweb.freebsd.org/changeset/base/301861 Log: MFC 295743,295744,295745,295746,295747 295743 hyperv/hn: Change global tunable prefix to hw.hn And use SYSCTL+CTLFLAG_RDTUN for them. Suggested by: adrian Reviewed by: adrian, Hongjiang Zhang <honzhan microsoft com> Approved by: adrian (mentor) MFC after: 1 week Sponsored by: Microsoft OSTC Differential Revision: https://reviews.freebsd.org/D5274 295744 hyperv/hn: Split RX ring data structure out of softc This paves the way for upcoming vRSS stuffs and eases more code cleanup. Reviewed by: adrian Approved by: adrian (mentor) MFC after: 1 week Sponsored by: Microsoft OSTC Differential Revision: https://reviews.freebsd.org/D5275 295745 hyperv/hn: Use taskqueue_enqueue() This also eases experiment on the non-fast taskqueue. Reviewed by: adrian, Jun Su <junsu microsoft com> Approved by: adrian (mentor) MFC after: 1 week Sponsored by: Microsoft OSTC Differential Revision: https://reviews.freebsd.org/D5276 295746 hyperv/hn: Use non-fast taskqueue for transmission Performance stays same; so no need to use fast taskqueue here. 295747 hyperv/hn: Split TX ring data structure out of softc This paves the way for upcoming vRSS stuffs and eases more code cleanup. Reviewed by: adrian Approved by: adrian (mentor) MFC after: 1 week Sponsored by: Microsoft OSTC Differential Revision: https://reviews.freebsd.org/D5283 Modified: stable/10/sys/dev/hyperv/netvsc/hv_net_vsc.h stable/10/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c Directory Properties: stable/10/ (props changed) Modified: stable/10/sys/dev/hyperv/netvsc/hv_net_vsc.h ============================================================================== --- stable/10/sys/dev/hyperv/netvsc/hv_net_vsc.h Mon Jun 13 07:03:00 2016 (r301860) +++ stable/10/sys/dev/hyperv/netvsc/hv_net_vsc.h Mon Jun 13 07:30:54 2016 (r301861) @@ -993,31 +993,25 @@ typedef struct { struct hn_txdesc; SLIST_HEAD(hn_txdesc_list, hn_txdesc); -/* - * Device-specific softc structure - */ -typedef struct hn_softc { - struct ifnet *hn_ifp; - struct arpcom arpcom; - struct ifmedia hn_media; - device_t hn_dev; - uint8_t hn_unit; - int hn_carrier; - int hn_if_flags; - struct mtx hn_lock; - int hn_initdone; - /* See hv_netvsc_drv_freebsd.c for rules on how to use */ - int temp_unusable; - struct hv_device *hn_dev_obj; - netvsc_dev *net_dev; +struct hn_rx_ring { + struct lro_ctrl hn_lro; - struct hn_txdesc *hn_txdesc; - bus_dma_tag_t hn_tx_data_dtag; - bus_dma_tag_t hn_tx_rndis_dtag; - int hn_tx_chimney_size; - int hn_tx_chimney_max; - uint64_t hn_csum_assist; + /* Trust csum verification on host side */ + int hn_trust_hcsum; /* HN_TRUST_HCSUM_ */ + + u_long hn_csum_ip; + u_long hn_csum_tcp; + u_long hn_csum_udp; + u_long hn_csum_trusted; + u_long hn_lro_tried; + u_long hn_small_pkts; +} __aligned(CACHE_LINE_SIZE); + +#define HN_TRUST_HCSUM_IP 0x0001 +#define HN_TRUST_HCSUM_TCP 0x0002 +#define HN_TRUST_HCSUM_UDP 0x0004 +struct hn_tx_ring { struct mtx hn_txlist_spin; struct hn_txdesc_list hn_txlist; int hn_txdesc_cnt; @@ -1025,32 +1019,57 @@ typedef struct hn_softc { int hn_txeof; int hn_sched_tx; - int hn_direct_tx_size; struct taskqueue *hn_tx_taskq; struct task hn_start_task; struct task hn_txeof_task; - struct lro_ctrl hn_lro; + struct mtx hn_tx_lock; + struct hn_softc *hn_sc; - /* Trust csum verification on host side */ - int hn_trust_hcsum; /* HN_TRUST_HCSUM_ */ + int hn_direct_tx_size; + int hn_tx_chimney_size; + bus_dma_tag_t hn_tx_data_dtag; + uint64_t hn_csum_assist; - u_long hn_csum_ip; - u_long hn_csum_tcp; - u_long hn_csum_udp; - u_long hn_csum_trusted; - u_long hn_lro_tried; - u_long hn_small_pkts; u_long hn_no_txdescs; u_long hn_send_failed; u_long hn_txdma_failed; u_long hn_tx_collapsed; u_long hn_tx_chimney; -} hn_softc_t; -#define HN_TRUST_HCSUM_IP 0x0001 -#define HN_TRUST_HCSUM_TCP 0x0002 -#define HN_TRUST_HCSUM_UDP 0x0004 + /* Rarely used stuffs */ + struct hn_txdesc *hn_txdesc; + bus_dma_tag_t hn_tx_rndis_dtag; + struct sysctl_oid *hn_tx_sysctl_tree; +} __aligned(CACHE_LINE_SIZE); + +/* + * Device-specific softc structure + */ +typedef struct hn_softc { + struct ifnet *hn_ifp; + struct arpcom arpcom; + struct ifmedia hn_media; + device_t hn_dev; + uint8_t hn_unit; + int hn_carrier; + int hn_if_flags; + struct mtx hn_lock; + int hn_initdone; + /* See hv_netvsc_drv_freebsd.c for rules on how to use */ + int temp_unusable; + struct hv_device *hn_dev_obj; + netvsc_dev *net_dev; + + int hn_rx_ring_cnt; + struct hn_rx_ring *hn_rx_ring; + + int hn_tx_ring_cnt; + struct hn_tx_ring *hn_tx_ring; + int hn_tx_chimney_max; + struct taskqueue *hn_tx_taskq; + struct sysctl_oid *hn_tx_sysctl_tree; +} hn_softc_t; /* * Externs Modified: stable/10/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c ============================================================================== --- stable/10/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c Mon Jun 13 07:03:00 2016 (r301860) +++ stable/10/sys/dev/hyperv/netvsc/hv_netvsc_drv_freebsd.c Mon Jun 13 07:30:54 2016 (r301861) @@ -153,7 +153,7 @@ __FBSDID("$FreeBSD$"); struct hn_txdesc { SLIST_ENTRY(hn_txdesc) link; struct mbuf *m; - struct hn_softc *sc; + struct hn_tx_ring *txr; int refs; uint32_t flags; /* HN_TXD_FLAG_ */ netvsc_packet netvsc_pkt; /* XXX to be removed */ @@ -193,7 +193,6 @@ struct hn_txdesc { #define NV_LOCK_INIT(_sc, _name) \ mtx_init(&(_sc)->hn_lock, _name, MTX_NETWORK_LOCK, MTX_DEF) #define NV_LOCK(_sc) mtx_lock(&(_sc)->hn_lock) -#define NV_TRYLOCK(_sc) mtx_trylock(&(_sc)->hn_lock) #define NV_LOCK_ASSERT(_sc) mtx_assert(&(_sc)->hn_lock, MA_OWNED) #define NV_UNLOCK(_sc) mtx_unlock(&(_sc)->hn_lock) #define NV_LOCK_DESTROY(_sc) mtx_destroy(&(_sc)->hn_lock) @@ -205,41 +204,57 @@ struct hn_txdesc { int hv_promisc_mode = 0; /* normal mode by default */ +SYSCTL_NODE(_hw, OID_AUTO, hn, CTLFLAG_RD, NULL, "Hyper-V network interface"); + /* Trust tcp segements verification on host side. */ static int hn_trust_hosttcp = 1; -TUNABLE_INT("dev.hn.trust_hosttcp", &hn_trust_hosttcp); +SYSCTL_INT(_hw_hn, OID_AUTO, trust_hosttcp, CTLFLAG_RDTUN, + &hn_trust_hosttcp, 0, + "Trust tcp segement verification on host side, " + "when csum info is missing (global setting)"); /* Trust udp datagrams verification on host side. */ static int hn_trust_hostudp = 1; -TUNABLE_INT("dev.hn.trust_hostudp", &hn_trust_hostudp); +SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostudp, CTLFLAG_RDTUN, + &hn_trust_hostudp, 0, + "Trust udp datagram verification on host side, " + "when csum info is missing (global setting)"); /* Trust ip packets verification on host side. */ static int hn_trust_hostip = 1; -TUNABLE_INT("dev.hn.trust_hostip", &hn_trust_hostip); +SYSCTL_INT(_hw_hn, OID_AUTO, trust_hostip, CTLFLAG_RDTUN, + &hn_trust_hostip, 0, + "Trust ip packet verification on host side, " + "when csum info is missing (global setting)"); #if __FreeBSD_version >= 1100045 /* Limit TSO burst size */ static int hn_tso_maxlen = 0; -TUNABLE_INT("dev.hn.tso_maxlen", &hn_tso_maxlen); +SYSCTL_INT(_hw_hn, OID_AUTO, tso_maxlen, CTLFLAG_RDTUN, + &hn_tso_maxlen, 0, "TSO burst limit"); #endif /* Limit chimney send size */ static int hn_tx_chimney_size = 0; -TUNABLE_INT("dev.hn.tx_chimney_size", &hn_tx_chimney_size); +SYSCTL_INT(_hw_hn, OID_AUTO, tx_chimney_size, CTLFLAG_RDTUN, + &hn_tx_chimney_size, 0, "Chimney send packet size limit"); /* Limit the size of packet for direct transmission */ static int hn_direct_tx_size = HN_DIRECT_TX_SIZE_DEF; -TUNABLE_INT("dev.hn.direct_tx_size", &hn_direct_tx_size); +SYSCTL_INT(_hw_hn, OID_AUTO, direct_tx_size, CTLFLAG_RDTUN, + &hn_direct_tx_size, 0, "Size of the packet for direct transmission"); #if defined(INET) || defined(INET6) #if __FreeBSD_version >= 1100095 static int hn_lro_entry_count = HN_LROENT_CNT_DEF; -TUNABLE_INT("dev.hn.lro_entry_count", &hn_lro_entry_count); +SYSCTL_INT(_hw_hn, OID_AUTO, lro_entry_count, CTLFLAG_RDTUN, + &hn_lro_entry_count, 0, "LRO entry count"); #endif #endif static int hn_share_tx_taskq = 0; -TUNABLE_INT("hw.hn.share_tx_taskq", &hn_share_tx_taskq); +SYSCTL_INT(_hw_hn, OID_AUTO, share_tx_taskq, CTLFLAG_RDTUN, + &hn_share_tx_taskq, 0, "Enable shared TX taskqueue"); static struct taskqueue *hn_tx_taskq; @@ -250,9 +265,9 @@ static void hn_stop(hn_softc_t *sc); static void hn_ifinit_locked(hn_softc_t *sc); static void hn_ifinit(void *xsc); static int hn_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data); -static int hn_start_locked(struct ifnet *ifp, int len); +static int hn_start_locked(struct hn_tx_ring *txr, int len); static void hn_start(struct ifnet *ifp); -static void hn_start_txeof(struct ifnet *ifp); +static void hn_start_txeof(struct hn_tx_ring *); static int hn_ifmedia_upd(struct ifnet *ifp); static void hn_ifmedia_sts(struct ifnet *ifp, struct ifmediareq *ifmr); #if __FreeBSD_version >= 1100099 @@ -261,12 +276,22 @@ static int hn_lro_ackcnt_sysctl(SYSCTL_H #endif static int hn_trust_hcsum_sysctl(SYSCTL_HANDLER_ARGS); static int hn_tx_chimney_size_sysctl(SYSCTL_HANDLER_ARGS); +static int hn_rx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS); +static int hn_rx_stat_u64_sysctl(SYSCTL_HANDLER_ARGS); +static int hn_tx_stat_ulong_sysctl(SYSCTL_HANDLER_ARGS); +static int hn_tx_conf_int_sysctl(SYSCTL_HANDLER_ARGS); static int hn_check_iplen(const struct mbuf *, int); -static int hn_create_tx_ring(struct hn_softc *sc); -static void hn_destroy_tx_ring(struct hn_softc *sc); +static int hn_create_tx_ring(struct hn_softc *, int); +static void hn_destroy_tx_ring(struct hn_tx_ring *); +static int hn_create_tx_data(struct hn_softc *); +static void hn_destroy_tx_data(struct hn_softc *); static void hn_start_taskfunc(void *xsc, int pending); static void hn_txeof_taskfunc(void *xsc, int pending); -static int hn_encap(struct hn_softc *, struct hn_txdesc *, struct mbuf **); +static void hn_stop_tx_tasks(struct hn_softc *); +static int hn_encap(struct hn_tx_ring *, struct hn_txdesc *, struct mbuf **); +static void hn_create_rx_data(struct hn_softc *sc); +static void hn_destroy_rx_data(struct hn_softc *sc); +static void hn_set_tx_chimney_size(struct hn_softc *, int); static int hn_ifmedia_upd(struct ifnet *ifp __unused) @@ -332,17 +357,10 @@ netvsc_attach(device_t dev) hn_softc_t *sc; int unit = device_get_unit(dev); struct ifnet *ifp = NULL; - struct sysctl_oid_list *child; - struct sysctl_ctx_list *ctx; int error; #if __FreeBSD_version >= 1100045 int tso_maxlen; #endif -#if defined(INET) || defined(INET6) -#if __FreeBSD_version >= 1100095 - int lroent_cnt; -#endif -#endif sc = device_get_softc(dev); if (sc == NULL) { @@ -352,29 +370,15 @@ netvsc_attach(device_t dev) bzero(sc, sizeof(hn_softc_t)); sc->hn_unit = unit; sc->hn_dev = dev; - sc->hn_direct_tx_size = hn_direct_tx_size; - if (hn_trust_hosttcp) - sc->hn_trust_hcsum |= HN_TRUST_HCSUM_TCP; - if (hn_trust_hostudp) - sc->hn_trust_hcsum |= HN_TRUST_HCSUM_UDP; - if (hn_trust_hostip) - sc->hn_trust_hcsum |= HN_TRUST_HCSUM_IP; if (hn_tx_taskq == NULL) { - sc->hn_tx_taskq = taskqueue_create_fast("hn_tx", M_WAITOK, + sc->hn_tx_taskq = taskqueue_create("hn_tx", M_WAITOK, taskqueue_thread_enqueue, &sc->hn_tx_taskq); taskqueue_start_threads(&sc->hn_tx_taskq, 1, PI_NET, "%s tx", device_get_nameunit(dev)); } else { sc->hn_tx_taskq = hn_tx_taskq; } - TASK_INIT(&sc->hn_start_task, 0, hn_start_taskfunc, sc); - TASK_INIT(&sc->hn_txeof_task, 0, hn_txeof_taskfunc, sc); - - error = hn_create_tx_ring(sc); - if (error) - goto failed; - NV_LOCK_INIT(sc, "NetVSCLock"); sc->hn_dev_obj = device_ctx; @@ -382,6 +386,12 @@ netvsc_attach(device_t dev) ifp = sc->hn_ifp = sc->arpcom.ac_ifp = if_alloc(IFT_ETHER); ifp->if_softc = sc; + error = hn_create_tx_data(sc); + if (error) + goto failed; + + hn_create_rx_data(sc); + if_initname(ifp, device_get_name(dev), device_get_unit(dev)); ifp->if_dunit = unit; ifp->if_dname = NETVSC_DEVNAME; @@ -412,12 +422,7 @@ netvsc_attach(device_t dev) ifp->if_capenable |= IFCAP_VLAN_HWTAGGING | IFCAP_VLAN_MTU | IFCAP_HWCSUM | IFCAP_TSO | IFCAP_LRO; - - if (hv_vmbus_protocal_version >= HV_VMBUS_VERSION_WIN8_1) - sc->hn_csum_assist = HN_CSUM_ASSIST; - else - sc->hn_csum_assist = HN_CSUM_ASSIST_WIN8; - ifp->if_hwassist = sc->hn_csum_assist | CSUM_TSO; + ifp->if_hwassist = sc->hn_tx_ring[0].hn_csum_assist | CSUM_TSO; error = hv_rf_on_device_add(device_ctx, &device_info); if (error) @@ -427,24 +432,6 @@ netvsc_attach(device_t dev) sc->hn_carrier = 1; } -#if defined(INET) || defined(INET6) -#if __FreeBSD_version >= 1100095 - lroent_cnt = hn_lro_entry_count; - if (lroent_cnt < TCP_LRO_ENTRIES) - lroent_cnt = TCP_LRO_ENTRIES; - tcp_lro_init_args(&sc->hn_lro, ifp, lroent_cnt, 0); - device_printf(dev, "LRO: entry count %d\n", lroent_cnt); -#else - tcp_lro_init(&sc->hn_lro); - /* Driver private LRO settings */ - sc->hn_lro.ifp = ifp; -#endif -#if __FreeBSD_version >= 1100099 - sc->hn_lro.lro_length_lim = HN_LRO_LENLIM_DEF; - sc->hn_lro.lro_ackcnt_lim = HN_LRO_ACKCNT_DEF; -#endif -#endif /* INET || INET6 */ - #if __FreeBSD_version >= 1100045 tso_maxlen = hn_tso_maxlen; if (tso_maxlen <= 0 || tso_maxlen > IP_MAXPACKET) @@ -464,134 +451,14 @@ netvsc_attach(device_t dev) #endif sc->hn_tx_chimney_max = sc->net_dev->send_section_size; - sc->hn_tx_chimney_size = sc->hn_tx_chimney_max; + hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max); if (hn_tx_chimney_size > 0 && hn_tx_chimney_size < sc->hn_tx_chimney_max) - sc->hn_tx_chimney_size = hn_tx_chimney_size; - - /* - * Always schedule transmission instead of trying - * to do direct transmission. This one gives the - * best performance so far. - */ - sc->hn_sched_tx = 1; - - ctx = device_get_sysctl_ctx(dev); - child = SYSCTL_CHILDREN(device_get_sysctl_tree(dev)); - - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_queued", - CTLFLAG_RW, &sc->hn_lro.lro_queued, 0, "LRO queued"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "lro_flushed", - CTLFLAG_RW, &sc->hn_lro.lro_flushed, 0, "LRO flushed"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "lro_tried", - CTLFLAG_RW, &sc->hn_lro_tried, "# of LRO tries"); -#if __FreeBSD_version >= 1100099 - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_length_lim", - CTLTYPE_UINT | CTLFLAG_RW, sc, 0, hn_lro_lenlim_sysctl, "IU", - "Max # of data bytes to be aggregated by LRO"); - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "lro_ackcnt_lim", - CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_lro_ackcnt_sysctl, "I", - "Max # of ACKs to be aggregated by LRO"); -#endif - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hosttcp", - CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_TCP, - hn_trust_hcsum_sysctl, "I", - "Trust tcp segement verification on host side, " - "when csum info is missing"); - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostudp", - CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_UDP, - hn_trust_hcsum_sysctl, "I", - "Trust udp datagram verification on host side, " - "when csum info is missing"); - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "trust_hostip", - CTLTYPE_INT | CTLFLAG_RW, sc, HN_TRUST_HCSUM_IP, - hn_trust_hcsum_sysctl, "I", - "Trust ip packet verification on host side, " - "when csum info is missing"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_ip", - CTLFLAG_RW, &sc->hn_csum_ip, "RXCSUM IP"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_tcp", - CTLFLAG_RW, &sc->hn_csum_tcp, "RXCSUM TCP"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_udp", - CTLFLAG_RW, &sc->hn_csum_udp, "RXCSUM UDP"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "csum_trusted", - CTLFLAG_RW, &sc->hn_csum_trusted, - "# of packets that we trust host's csum verification"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "small_pkts", - CTLFLAG_RW, &sc->hn_small_pkts, "# of small packets received"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "no_txdescs", - CTLFLAG_RW, &sc->hn_no_txdescs, "# of times short of TX descs"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "send_failed", - CTLFLAG_RW, &sc->hn_send_failed, "# of hyper-v sending failure"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "txdma_failed", - CTLFLAG_RW, &sc->hn_txdma_failed, "# of TX DMA failure"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_collapsed", - CTLFLAG_RW, &sc->hn_tx_collapsed, "# of TX mbuf collapsed"); - SYSCTL_ADD_ULONG(ctx, child, OID_AUTO, "tx_chimney", - CTLFLAG_RW, &sc->hn_tx_chimney, "# of chimney send"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_cnt", - CTLFLAG_RD, &sc->hn_txdesc_cnt, 0, "# of total TX descs"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "txdesc_avail", - CTLFLAG_RD, &sc->hn_txdesc_avail, 0, "# of available TX descs"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "tx_chimney_max", - CTLFLAG_RD, &sc->hn_tx_chimney_max, 0, - "Chimney send packet size upper boundary"); - SYSCTL_ADD_PROC(ctx, child, OID_AUTO, "tx_chimney_size", - CTLTYPE_INT | CTLFLAG_RW, sc, 0, hn_tx_chimney_size_sysctl, - "I", "Chimney send packet size limit"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "direct_tx_size", - CTLFLAG_RW, &sc->hn_direct_tx_size, 0, - "Size of the packet for direct transmission"); - SYSCTL_ADD_INT(ctx, child, OID_AUTO, "sched_tx", - CTLFLAG_RW, &sc->hn_sched_tx, 0, - "Always schedule transmission " - "instead of doing direct transmission"); - - if (unit == 0) { - struct sysctl_ctx_list *dc_ctx; - struct sysctl_oid_list *dc_child; - devclass_t dc; - - /* - * Add sysctl nodes for devclass - */ - dc = device_get_devclass(dev); - dc_ctx = devclass_get_sysctl_ctx(dc); - dc_child = SYSCTL_CHILDREN(devclass_get_sysctl_tree(dc)); - - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hosttcp", - CTLFLAG_RD, &hn_trust_hosttcp, 0, - "Trust tcp segement verification on host side, " - "when csum info is missing (global setting)"); - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostudp", - CTLFLAG_RD, &hn_trust_hostudp, 0, - "Trust udp datagram verification on host side, " - "when csum info is missing (global setting)"); - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "trust_hostip", - CTLFLAG_RD, &hn_trust_hostip, 0, - "Trust ip packet verification on host side, " - "when csum info is missing (global setting)"); - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tx_chimney_size", - CTLFLAG_RD, &hn_tx_chimney_size, 0, - "Chimney send packet size limit"); -#if __FreeBSD_version >= 1100045 - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "tso_maxlen", - CTLFLAG_RD, &hn_tso_maxlen, 0, "TSO burst limit"); -#endif - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "direct_tx_size", - CTLFLAG_RD, &hn_direct_tx_size, 0, - "Size of the packet for direct transmission"); -#if defined(INET) || defined(INET6) -#if __FreeBSD_version >= 1100095 - SYSCTL_ADD_INT(dc_ctx, dc_child, OID_AUTO, "lro_entry_count", - CTLFLAG_RD, &hn_lro_entry_count, 0, "LRO entry count"); -#endif -#endif - } + hn_set_tx_chimney_size(sc, hn_tx_chimney_size); return (0); failed: - hn_destroy_tx_ring(sc); + hn_destroy_tx_data(sc); if (ifp != NULL) if_free(ifp); return (error); @@ -622,16 +489,14 @@ netvsc_detach(device_t dev) hv_rf_on_device_remove(hv_device, HV_RF_NV_DESTROY_CHANNEL); - taskqueue_drain(sc->hn_tx_taskq, &sc->hn_start_task); - taskqueue_drain(sc->hn_tx_taskq, &sc->hn_txeof_task); - if (sc->hn_tx_taskq != hn_tx_taskq) - taskqueue_free(sc->hn_tx_taskq); + hn_stop_tx_tasks(sc); ifmedia_removeall(&sc->hn_media); -#if defined(INET) || defined(INET6) - tcp_lro_free(&sc->hn_lro); -#endif - hn_destroy_tx_ring(sc); + hn_destroy_rx_data(sc); + hn_destroy_tx_data(sc); + + if (sc->hn_tx_taskq != hn_tx_taskq) + taskqueue_free(sc->hn_tx_taskq); return (0); } @@ -646,13 +511,13 @@ netvsc_shutdown(device_t dev) } static __inline int -hn_txdesc_dmamap_load(struct hn_softc *sc, struct hn_txdesc *txd, +hn_txdesc_dmamap_load(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head, bus_dma_segment_t *segs, int *nsegs) { struct mbuf *m = *m_head; int error; - error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag, txd->data_dmap, + error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT); if (error == EFBIG) { struct mbuf *m_new; @@ -662,13 +527,13 @@ hn_txdesc_dmamap_load(struct hn_softc *s return ENOBUFS; else *m_head = m = m_new; - sc->hn_tx_collapsed++; + txr->hn_tx_collapsed++; - error = bus_dmamap_load_mbuf_sg(sc->hn_tx_data_dtag, + error = bus_dmamap_load_mbuf_sg(txr->hn_tx_data_dtag, txd->data_dmap, m, segs, nsegs, BUS_DMA_NOWAIT); } if (!error) { - bus_dmamap_sync(sc->hn_tx_data_dtag, txd->data_dmap, + bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap, BUS_DMASYNC_PREWRITE); txd->flags |= HN_TXD_FLAG_DMAMAP; } @@ -676,20 +541,20 @@ hn_txdesc_dmamap_load(struct hn_softc *s } static __inline void -hn_txdesc_dmamap_unload(struct hn_softc *sc, struct hn_txdesc *txd) +hn_txdesc_dmamap_unload(struct hn_tx_ring *txr, struct hn_txdesc *txd) { if (txd->flags & HN_TXD_FLAG_DMAMAP) { - bus_dmamap_sync(sc->hn_tx_data_dtag, + bus_dmamap_sync(txr->hn_tx_data_dtag, txd->data_dmap, BUS_DMASYNC_POSTWRITE); - bus_dmamap_unload(sc->hn_tx_data_dtag, + bus_dmamap_unload(txr->hn_tx_data_dtag, txd->data_dmap); txd->flags &= ~HN_TXD_FLAG_DMAMAP; } } static __inline int -hn_txdesc_put(struct hn_softc *sc, struct hn_txdesc *txd) +hn_txdesc_put(struct hn_tx_ring *txr, struct hn_txdesc *txd) { KASSERT((txd->flags & HN_TXD_FLAG_ONLIST) == 0, @@ -699,7 +564,7 @@ hn_txdesc_put(struct hn_softc *sc, struc if (atomic_fetchadd_int(&txd->refs, -1) != 1) return 0; - hn_txdesc_dmamap_unload(sc, txd); + hn_txdesc_dmamap_unload(txr, txd); if (txd->m != NULL) { m_freem(txd->m); txd->m = NULL; @@ -707,31 +572,31 @@ hn_txdesc_put(struct hn_softc *sc, struc txd->flags |= HN_TXD_FLAG_ONLIST; - mtx_lock_spin(&sc->hn_txlist_spin); - KASSERT(sc->hn_txdesc_avail >= 0 && - sc->hn_txdesc_avail < sc->hn_txdesc_cnt, - ("txdesc_put: invalid txd avail %d", sc->hn_txdesc_avail)); - sc->hn_txdesc_avail++; - SLIST_INSERT_HEAD(&sc->hn_txlist, txd, link); - mtx_unlock_spin(&sc->hn_txlist_spin); + mtx_lock_spin(&txr->hn_txlist_spin); + KASSERT(txr->hn_txdesc_avail >= 0 && + txr->hn_txdesc_avail < txr->hn_txdesc_cnt, + ("txdesc_put: invalid txd avail %d", txr->hn_txdesc_avail)); + txr->hn_txdesc_avail++; + SLIST_INSERT_HEAD(&txr->hn_txlist, txd, link); + mtx_unlock_spin(&txr->hn_txlist_spin); return 1; } static __inline struct hn_txdesc * -hn_txdesc_get(struct hn_softc *sc) +hn_txdesc_get(struct hn_tx_ring *txr) { struct hn_txdesc *txd; - mtx_lock_spin(&sc->hn_txlist_spin); - txd = SLIST_FIRST(&sc->hn_txlist); + mtx_lock_spin(&txr->hn_txlist_spin); + txd = SLIST_FIRST(&txr->hn_txlist); if (txd != NULL) { - KASSERT(sc->hn_txdesc_avail > 0, - ("txdesc_get: invalid txd avail %d", sc->hn_txdesc_avail)); - sc->hn_txdesc_avail--; - SLIST_REMOVE_HEAD(&sc->hn_txlist, link); + KASSERT(txr->hn_txdesc_avail > 0, + ("txdesc_get: invalid txd avail %d", txr->hn_txdesc_avail)); + txr->hn_txdesc_avail--; + SLIST_REMOVE_HEAD(&txr->hn_txlist, link); } - mtx_unlock_spin(&sc->hn_txlist_spin); + mtx_unlock_spin(&txr->hn_txlist_spin); if (txd != NULL) { KASSERT(txd->m == NULL && txd->refs == 0 && @@ -763,22 +628,24 @@ netvsc_xmit_completion(void *context) { netvsc_packet *packet = context; struct hn_txdesc *txd; - struct hn_softc *sc; + struct hn_tx_ring *txr; txd = (struct hn_txdesc *)(uintptr_t) packet->compl.send.send_completion_tid; - sc = txd->sc; - sc->hn_txeof = 1; - hn_txdesc_put(sc, txd); + txr = txd->txr; + txr->hn_txeof = 1; + hn_txdesc_put(txr, txd); } void netvsc_channel_rollup(struct hv_device *device_ctx) { struct hn_softc *sc = device_get_softc(device_ctx->device); + struct hn_tx_ring *txr = &sc->hn_tx_ring[0]; /* TODO: vRSS */ #if defined(INET) || defined(INET6) - struct lro_ctrl *lro = &sc->hn_lro; + struct hn_rx_ring *rxr = &sc->hn_rx_ring[0]; /* TODO: vRSS */ + struct lro_ctrl *lro = &rxr->hn_lro; struct lro_entry *queued; while ((queued = SLIST_FIRST(&lro->lro_active)) != NULL) { @@ -787,11 +654,11 @@ netvsc_channel_rollup(struct hv_device * } #endif - if (!sc->hn_txeof) + if (!txr->hn_txeof) return; - sc->hn_txeof = 0; - hn_start_txeof(sc->hn_ifp); + txr->hn_txeof = 0; + hn_start_txeof(txr); } /* @@ -799,7 +666,7 @@ netvsc_channel_rollup(struct hv_device * * If this function fails, then both txd and m_head0 will be freed. */ static int -hn_encap(struct hn_softc *sc, struct hn_txdesc *txd, struct mbuf **m_head0) +hn_encap(struct hn_tx_ring *txr, struct hn_txdesc *txd, struct mbuf **m_head0) { bus_dma_segment_t segs[HN_TX_DATA_SEGCNT_MAX]; int error, nsegs, i; @@ -902,7 +769,7 @@ hn_encap(struct hn_softc *sc, struct hn_ #endif tso_info->lso_v2_xmit.tcp_header_offset = 0; tso_info->lso_v2_xmit.mss = m_head->m_pkthdr.tso_segsz; - } else if (m_head->m_pkthdr.csum_flags & sc->hn_csum_assist) { + } else if (m_head->m_pkthdr.csum_flags & txr->hn_csum_assist) { rndis_tcp_ip_csum_info *csum_info; rndis_msg_size += RNDIS_CSUM_PPI_SIZE; @@ -929,8 +796,8 @@ hn_encap(struct hn_softc *sc, struct hn_ /* * Chimney send, if the packet could fit into one chimney buffer. */ - if (packet->tot_data_buf_len < sc->hn_tx_chimney_size) { - netvsc_dev *net_dev = sc->net_dev; + if (packet->tot_data_buf_len < txr->hn_tx_chimney_size) { + netvsc_dev *net_dev = txr->hn_sc->net_dev; uint32_t send_buf_section_idx; send_buf_section_idx = @@ -949,12 +816,12 @@ hn_encap(struct hn_softc *sc, struct hn_ packet->send_buf_section_size = packet->tot_data_buf_len; packet->page_buf_count = 0; - sc->hn_tx_chimney++; + txr->hn_tx_chimney++; goto done; } } - error = hn_txdesc_dmamap_load(sc, txd, &m_head, segs, &nsegs); + error = hn_txdesc_dmamap_load(txr, txd, &m_head, segs, &nsegs); if (error) { int freed; @@ -964,12 +831,12 @@ hn_encap(struct hn_softc *sc, struct hn_ m_freem(m_head); *m_head0 = NULL; - freed = hn_txdesc_put(sc, txd); + freed = hn_txdesc_put(txr, txd); KASSERT(freed != 0, ("fail to free txd upon txdma error")); - sc->hn_txdma_failed++; - if_inc_counter(sc->hn_ifp, IFCOUNTER_OERRORS, 1); + txr->hn_txdma_failed++; + if_inc_counter(txr->hn_sc->hn_ifp, IFCOUNTER_OERRORS, 1); return error; } *m_head0 = m_head; @@ -1012,11 +879,15 @@ done: * Start a transmit of one or more packets */ static int -hn_start_locked(struct ifnet *ifp, int len) +hn_start_locked(struct hn_tx_ring *txr, int len) { - struct hn_softc *sc = ifp->if_softc; + struct hn_softc *sc = txr->hn_sc; + struct ifnet *ifp = sc->hn_ifp; struct hv_device *device_ctx = vmbus_get_devctx(sc->hn_dev); + KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring")); + mtx_assert(&txr->hn_tx_lock, MA_OWNED); + if ((ifp->if_drv_flags & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING) return 0; @@ -1040,15 +911,15 @@ hn_start_locked(struct ifnet *ifp, int l return 1; } - txd = hn_txdesc_get(sc); + txd = hn_txdesc_get(txr); if (txd == NULL) { - sc->hn_no_txdescs++; + txr->hn_no_txdescs++; IF_PREPEND(&ifp->if_snd, m_head); atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); break; } - error = hn_encap(sc, txd, &m_head); + error = hn_encap(txr, txd, &m_head); if (error) { /* Both txd and m_head are freed */ continue; @@ -1063,7 +934,7 @@ again: ETHER_BPF_MTAP(ifp, m_head); if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); } - hn_txdesc_put(sc, txd); + hn_txdesc_put(txr, txd); if (__predict_false(error)) { int freed; @@ -1075,9 +946,9 @@ again: * commands to run? Ask netvsc_channel_rollup() * to kick start later. */ - sc->hn_txeof = 1; + txr->hn_txeof = 1; if (!send_failed) { - sc->hn_send_failed++; + txr->hn_send_failed++; send_failed = 1; /* * Try sending again after set hn_txeof; @@ -1094,11 +965,11 @@ again: * DMA map in hn_txdesc_put(), if it was loaded. */ txd->m = NULL; - freed = hn_txdesc_put(sc, txd); + freed = hn_txdesc_put(txr, txd); KASSERT(freed != 0, ("fail to free txd upon send error")); - sc->hn_send_failed++; + txr->hn_send_failed++; IF_PREPEND(&ifp->if_snd, m_head); atomic_set_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); break; @@ -1194,10 +1065,10 @@ int netvsc_recv(struct hv_device *device_ctx, netvsc_packet *packet, rndis_tcp_ip_csum_info *csum_info) { - hn_softc_t *sc = (hn_softc_t *)device_get_softc(device_ctx->device); + struct hn_softc *sc = device_get_softc(device_ctx->device); + struct hn_rx_ring *rxr = &sc->hn_rx_ring[0]; /* TODO: vRSS */ struct mbuf *m_new; struct ifnet *ifp; - device_t dev = device_ctx->device; int size, do_lro = 0, do_csum = 1; if (sc == NULL) { @@ -1224,7 +1095,7 @@ netvsc_recv(struct hv_device *device_ctx memcpy(mtod(m_new, void *), packet->data, packet->tot_data_buf_len); m_new->m_pkthdr.len = m_new->m_len = packet->tot_data_buf_len; - sc->hn_small_pkts++; + rxr->hn_small_pkts++; } else { /* * Get an mbuf with a cluster. For packets 2K or less, @@ -1240,7 +1111,7 @@ netvsc_recv(struct hv_device *device_ctx m_new = m_getjcl(M_NOWAIT, MT_DATA, M_PKTHDR, size); if (m_new == NULL) { - device_printf(dev, "alloc mbuf failed.\n"); + if_printf(ifp, "alloc mbuf failed.\n"); return (0); } @@ -1257,7 +1128,7 @@ netvsc_recv(struct hv_device *device_ctx if (csum_info->receive.ip_csum_succeeded && do_csum) { m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); - sc->hn_csum_ip++; + rxr->hn_csum_ip++; } /* TCP/UDP csum offload */ @@ -1267,9 +1138,9 @@ netvsc_recv(struct hv_device *device_ctx (CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m_new->m_pkthdr.csum_data = 0xffff; if (csum_info->receive.tcp_csum_succeeded) - sc->hn_csum_tcp++; + rxr->hn_csum_tcp++; else - sc->hn_csum_udp++; + rxr->hn_csum_udp++; } if (csum_info->receive.ip_csum_succeeded && @@ -1301,8 +1172,9 @@ netvsc_recv(struct hv_device *device_ctx pr = hn_check_iplen(m_new, hoff); if (pr == IPPROTO_TCP) { if (do_csum && - (sc->hn_trust_hcsum & HN_TRUST_HCSUM_TCP)) { - sc->hn_csum_trusted++; + (rxr->hn_trust_hcsum & + HN_TRUST_HCSUM_TCP)) { + rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID | CSUM_PSEUDO_HDR); @@ -1312,16 +1184,17 @@ netvsc_recv(struct hv_device *device_ctx do_lro = 1; } else if (pr == IPPROTO_UDP) { if (do_csum && - (sc->hn_trust_hcsum & HN_TRUST_HCSUM_UDP)) { - sc->hn_csum_trusted++; + (rxr->hn_trust_hcsum & + HN_TRUST_HCSUM_UDP)) { + rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID | CSUM_DATA_VALID | CSUM_PSEUDO_HDR); m_new->m_pkthdr.csum_data = 0xffff; } } else if (pr != IPPROTO_DONE && do_csum && - (sc->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) { - sc->hn_csum_trusted++; + (rxr->hn_trust_hcsum & HN_TRUST_HCSUM_IP)) { + rxr->hn_csum_trusted++; m_new->m_pkthdr.csum_flags |= (CSUM_IP_CHECKED | CSUM_IP_VALID); } @@ -1343,10 +1216,10 @@ skip: if ((ifp->if_capenable & IFCAP_LRO) && do_lro) { #if defined(INET) || defined(INET6) - struct lro_ctrl *lro = &sc->hn_lro; + struct lro_ctrl *lro = &rxr->hn_lro; if (lro->lro_cnt) { - sc->hn_lro_tried++; + rxr->hn_lro_tried++; if (tcp_lro_rx(lro, m_new, 0) == 0) { /* DONE! */ return 0; @@ -1427,8 +1300,16 @@ hn_ioctl(struct ifnet *ifp, u_long cmd, * Make sure that LRO aggregation length limit is still * valid, after the MTU change. */ - if (sc->hn_lro.lro_length_lim < HN_LRO_LENLIM_MIN(ifp)) - sc->hn_lro.lro_length_lim = HN_LRO_LENLIM_MIN(ifp); + NV_LOCK(sc); + if (sc->hn_rx_ring[0].hn_lro.lro_length_lim < + HN_LRO_LENLIM_MIN(ifp)) { + int i; + for (i = 0; i < sc->hn_rx_ring_cnt; ++i) { + sc->hn_rx_ring[i].hn_lro.lro_length_lim = + HN_LRO_LENLIM_MIN(ifp); + } + } + NV_UNLOCK(sc); #endif do { @@ -1469,8 +1350,10 @@ hn_ioctl(struct ifnet *ifp, u_long cmd, } sc->hn_tx_chimney_max = sc->net_dev->send_section_size; - if (sc->hn_tx_chimney_size > sc->hn_tx_chimney_max) - sc->hn_tx_chimney_size = sc->hn_tx_chimney_max; + if (sc->hn_tx_ring[0].hn_tx_chimney_size > + sc->hn_tx_chimney_max) + hn_set_tx_chimney_size(sc, sc->hn_tx_chimney_max); + hn_ifinit_locked(sc); NV_LOCK(sc); @@ -1535,10 +1418,13 @@ hn_ioctl(struct ifnet *ifp, u_long cmd, mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (mask & IFCAP_TXCSUM) { ifp->if_capenable ^= IFCAP_TXCSUM; - if (ifp->if_capenable & IFCAP_TXCSUM) - ifp->if_hwassist |= sc->hn_csum_assist; - else - ifp->if_hwassist &= ~sc->hn_csum_assist; + if (ifp->if_capenable & IFCAP_TXCSUM) { + ifp->if_hwassist |= + sc->hn_tx_ring[0].hn_csum_assist; + } else { + ifp->if_hwassist &= + ~sc->hn_tx_ring[0].hn_csum_assist; + } } if (mask & IFCAP_RXCSUM) @@ -1621,50 +1507,54 @@ static void hn_start(struct ifnet *ifp) { struct hn_softc *sc = ifp->if_softc; + struct hn_tx_ring *txr = &sc->hn_tx_ring[0]; - if (sc->hn_sched_tx) + if (txr->hn_sched_tx) goto do_sched; - if (NV_TRYLOCK(sc)) { + if (mtx_trylock(&txr->hn_tx_lock)) { int sched; - sched = hn_start_locked(ifp, sc->hn_direct_tx_size); - NV_UNLOCK(sc); + sched = hn_start_locked(txr, txr->hn_direct_tx_size); + mtx_unlock(&txr->hn_tx_lock); if (!sched) return; } do_sched: - taskqueue_enqueue_fast(sc->hn_tx_taskq, &sc->hn_start_task); + taskqueue_enqueue(txr->hn_tx_taskq, &txr->hn_start_task); } static void -hn_start_txeof(struct ifnet *ifp) +hn_start_txeof(struct hn_tx_ring *txr) { - struct hn_softc *sc = ifp->if_softc; + struct hn_softc *sc = txr->hn_sc; + struct ifnet *ifp = sc->hn_ifp; - if (sc->hn_sched_tx) + KASSERT(txr == &sc->hn_tx_ring[0], ("not the first TX ring")); + + if (txr->hn_sched_tx) goto do_sched; - if (NV_TRYLOCK(sc)) { + if (mtx_trylock(&txr->hn_tx_lock)) { int sched; atomic_clear_int(&ifp->if_drv_flags, IFF_DRV_OACTIVE); - sched = hn_start_locked(ifp, sc->hn_direct_tx_size); - NV_UNLOCK(sc); + sched = hn_start_locked(txr, txr->hn_direct_tx_size); + mtx_unlock(&txr->hn_tx_lock); if (sched) { - taskqueue_enqueue_fast(sc->hn_tx_taskq, - &sc->hn_start_task); *** DIFF OUTPUT TRUNCATED AT 1000 LINES ***
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?201606130730.u5D7Utc6086560>