[PATCH V3 0/5] Add LE hash collision bug fix for active and passive offloaded connections
This patch series fixes the LE hash collision issue in cxgb4 and RDMA/cxgb4 drivers in kernel.org. If the hash functionality is enabled in T4 then tuple information of active and passive offloaded connections are stored in DDR3 memory. LE (Lookup Engine) implements the interface to search this tuple entries using hash algorithm. To avoid LE hash collision, firmware provides new mechanisms to setup active and passive open connections. In case of active open connection, firmware detects LE hash collision situation and notifies driver. Driver uses fw_ofld_connection work request to offload that connection. Its tuple information will be stored in TCAM memory array. In case of passive open connection, server filter region is created in TCAM. This region stores the filter which will redirect the incoming SYN packet to offload queues. After this driver tries to establish the connection using firmware work request. This patch series also adds framework for managing filters and to use T4's filter capabilities. Following testing has been carried out successfully on this patch series. 1. 1000 active open connections created and saw that tcam_full counter is getting incremented through debugfs file. 2. Multiple passive open connections were created and ran the same test repetatively to verify server filter is getting created and deleted properly. The patch-series is based on Roland's infiniband tree (for-next branch), and involves changes on cxgb4 and RDMA/cxgb4 drivers. Both linux-rdma and netdev are included in this post for review. We have some more bug fixes in RDMA/cxgb4 after this patch series. So, we would like to request this series to get mereged through Roland's infiniband for-next tree. V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 Thanks, Vipul Pandya Vipul Pandya (5): cxgb4: Add T4 filter support cxgb4: Add LE hash collision bug fix path in LLD driver RDMA/cxgb4: Fix LE hash collision bug for active open connection RDMA/cxgb4: Fix LE hash collision bug for passive open connection RDMA/cxgb4: Fix bug for active and passive LE hash collision path drivers/infiniband/hw/cxgb4/cm.c| 789 +++--- drivers/infiniband/hw/cxgb4/device.c| 210 ++- drivers/infiniband/hw/cxgb4/iw_cxgb4.h | 33 + drivers/net/ethernet/chelsio/cxgb4/cxgb4.h | 146 + drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 473 +- drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h | 23 +- drivers/net/ethernet/chelsio/cxgb4/l2t.c| 34 + drivers/net/ethernet/chelsio/cxgb4/l2t.h|3 + drivers/net/ethernet/chelsio/cxgb4/t4_hw.c | 23 +- drivers/net/ethernet/chelsio/cxgb4/t4_msg.h | 66 ++ drivers/net/ethernet/chelsio/cxgb4/t4_regs.h| 37 ++ drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h | 388 +++ 12 files changed, 2100 insertions(+), 125 deletions(-) -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH V3 1/5] cxgb4: Add T4 filter support
The T4 architecture is capable of filtering ingress packets at line rate using the rule in TCAM. If packet hits a rule in the TCAM then it can be either dropped or passed to the receive queues based on a rule settings. This patch adds framework for managing filters and to use T4's filter capabilities. It constructs a Firmware Filter Work Request which writes the filter at a specified index to get the work done. It hosts shadow copy of ingress filter entry to check field size limitations and save memory in the case where the filter table is large. Signed-off-by: Vipul Pandya vi...@chelsio.com --- V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h | 134 ++ drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 301 ++- drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h |2 + drivers/net/ethernet/chelsio/cxgb4/l2t.c| 33 +++ drivers/net/ethernet/chelsio/cxgb4/l2t.h|3 + drivers/net/ethernet/chelsio/cxgb4/t4_hw.c | 23 ++- drivers/net/ethernet/chelsio/cxgb4/t4_msg.h |1 + drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h | 279 + 8 files changed, 770 insertions(+), 6 deletions(-) diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h index 378988b..b6403d1 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h @@ -545,6 +545,132 @@ struct adapter { spinlock_t stats_lock; }; +/* + * Defined bit width of user definable filter tuples + */ +#define ETHTYPE_BITWIDTH 16 +#define FRAG_BITWIDTH 1 +#define MACIDX_BITWIDTH 9 +#define FCOE_BITWIDTH 1 +#define IPORT_BITWIDTH 3 +#define MATCHTYPE_BITWIDTH 3 +#define PROTO_BITWIDTH 8 +#define TOS_BITWIDTH 8 +#define PF_BITWIDTH 8 +#define VF_BITWIDTH 8 +#define IVLAN_BITWIDTH 16 +#define OVLAN_BITWIDTH 16 + +/* Filter matching rules. These consist of a set of ingress packet field + * (value, mask) tuples. The associated ingress packet field matches the + * tuple when ((field mask) == value). (Thus a wildcard don't care field + * rule can be constructed by specifying a tuple of (0, 0).) A filter rule + * matches an ingress packet when all of the individual individual field + * matching rules are true. + * + * Partial field masks are always valid, however, while it may be easy to + * understand their meanings for some fields (e.g. IP address to match a + * subnet), for others making sensible partial masks is less intuitive (e.g. + * MPS match type) ... + * + * Most of the following data structures are modeled on T4 capabilities. + * Drivers for earlier chips use the subsets which make sense for those chips. + * We really need to come up with a hardware-independent mechanism to + * represent hardware filter capabilities ... + */ +struct ch_filter_tuple { + /* +* Compressed header matching field rules. The TP_VLAN_PRI_MAP +* register selects which of these fields will participate in the +* filter match rules -- up to a maximum of 36 bits. Because +* TP_VLAN_PRI_MAP is a global register, all filters must use the same +* set of fields. +*/ + uint32_t ethtype:ETHTYPE_BITWIDTH; /* Ethernet type */ + uint32_t frag:FRAG_BITWIDTH;/* IP fragmentation header */ + uint32_t ivlan_vld:1; /* inner VLAN valid */ + uint32_t ovlan_vld:1; /* outer VLAN valid */ + uint32_t pfvf_vld:1;/* PF/VF valid */ + uint32_t macidx:MACIDX_BITWIDTH;/* exact match MAC index */ + uint32_t fcoe:FCOE_BITWIDTH;/* FCoE packet */ + uint32_t iport:IPORT_BITWIDTH; /* ingress port */ + uint32_t matchtype:MATCHTYPE_BITWIDTH; /* MPS match type */ + uint32_t proto:PROTO_BITWIDTH; /* protocol type */ + uint32_t tos:TOS_BITWIDTH; /* TOS/Traffic Type */ + uint32_t pf:PF_BITWIDTH;/* PCI-E PF ID */ + uint32_t vf:VF_BITWIDTH;/* PCI-E VF ID */ + uint32_t ivlan:IVLAN_BITWIDTH; /* inner VLAN */ + uint32_t ovlan:OVLAN_BITWIDTH; /* outer VLAN */ + + /* Uncompressed header matching field rules. These are always +* available for field rules. +*/ + uint8_t lip[16];/* local IP address (IPv4 in [3:0]) */ + uint8_t fip[16];/* foreign IP address (IPv4 in [3:0]) */ + uint16_t lport; /* local port */ + uint16_t fport; /* foreign port */ +}; + +/* A filter ioctl command. + */ +struct ch_filter_specification { + /* +* Administrative fields for filter. +*/ + uint32_t hitcnts:1; /* count filter hits in TCB */ + uint32_t prio:1;/* filter has priority over active/server */ +
[PATCH V3 2/5] cxgb4: Add LE hash collision bug fix path in LLD driver
It supports establishing passive open connection through firmware filter work request. Passive open connection will go through this path as now instead of listening server we create a server filter which will redirect the incoming SYN packet to the offload queue. It divides filter region into regular filters and server filter portion. It introduces new server filter region which will be exclusively used for creating server filters. This region will not overlap with regular filter region. It provides new API cxgb4_alloc_sftid in LLD for getting stid in case of LE hash collision path. This new stid will be used to open server filter in the filter region. Signed-off-by: Vipul Pandya vi...@chelsio.com --- V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 drivers/net/ethernet/chelsio/cxgb4/cxgb4.h |4 + drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 150 +- drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h | 15 ++- drivers/net/ethernet/chelsio/cxgb4/t4_regs.h| 33 + 4 files changed, 193 insertions(+), 9 deletions(-) diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h index b6403d1..868acde 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4.h @@ -35,6 +35,8 @@ #ifndef __CXGB4_H__ #define __CXGB4_H__ +#include t4_hw.h + #include linux/bitops.h #include linux/cache.h #include linux/interrupt.h @@ -212,6 +214,8 @@ struct tp_err_stats { struct tp_params { unsigned int ntxchan;/* # of Tx channels */ unsigned int tre;/* log2 of core clocks per TP tick */ + unsigned short tx_modq_map; /* TX modulation scheduler queue to */ +/* channel map */ uint32_t dack_re;/* DACK timer resolution */ unsigned short tx_modq[NCHAN]; /* channel to modulation queue map */ diff --git a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c index a678287..f55386b 100644 --- a/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c +++ b/drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c @@ -2489,7 +2489,34 @@ int cxgb4_alloc_stid(struct tid_info *t, int family, void *data) EXPORT_SYMBOL(cxgb4_alloc_stid); /* - * Release a server TID. + * Allocate a server filter TID and set it to the supplied value. + */ +int cxgb4_alloc_sftid(struct tid_info *t, int family, void *data) +{ + int stid; + + spin_lock_bh(t-stid_lock); + if (family == PF_INET) { + stid = find_next_zero_bit(t-stid_bmap, + t-nstids + t-nsftids, t-nstids); + if (stid (t-nstids + t-nsftids)) + __set_bit(stid, t-stid_bmap); + else + stid = -1; + } else { + stid = -1; + } + if (stid = 0) { + t-stid_tab[stid].data = data; + stid += t-stid_base; + t-stids_in_use++; + } + spin_unlock_bh(t-stid_lock); + return stid; +} +EXPORT_SYMBOL(cxgb4_alloc_sftid); + +/* Release a server TID. */ void cxgb4_free_stid(struct tid_info *t, unsigned int stid, int family) { @@ -2604,12 +2631,14 @@ static int tid_init(struct tid_info *t) unsigned int stid_bmap_size; unsigned int natids = t-natids; - stid_bmap_size = BITS_TO_LONGS(t-nstids); + stid_bmap_size = BITS_TO_LONGS(t-nstids + t-nsftids); size = t-ntids * sizeof(*t-tid_tab) + natids * sizeof(*t-atid_tab) + t-nstids * sizeof(*t-stid_tab) + + t-nsftids * sizeof(*t-stid_tab) + stid_bmap_size * sizeof(long) + - t-nftids * sizeof(*t-ftid_tab); + t-nftids * sizeof(*t-ftid_tab) + + t-nsftids * sizeof(*t-ftid_tab); t-tid_tab = t4_alloc_mem(size); if (!t-tid_tab) @@ -2617,7 +2646,7 @@ static int tid_init(struct tid_info *t) t-atid_tab = (union aopen_entry *)t-tid_tab[t-ntids]; t-stid_tab = (struct serv_entry *)t-atid_tab[natids]; - t-stid_bmap = (unsigned long *)t-stid_tab[t-nstids]; + t-stid_bmap = (unsigned long *)t-stid_tab[t-nstids + t-nsftids]; t-ftid_tab = (struct filter_entry *)t-stid_bmap[stid_bmap_size]; spin_lock_init(t-stid_lock); spin_lock_init(t-atid_lock); @@ -2633,7 +2662,7 @@ static int tid_init(struct tid_info *t) t-atid_tab[natids - 1].next = t-atid_tab[natids]; t-afree = t-atid_tab; } - bitmap_zero(t-stid_bmap, t-nstids); + bitmap_zero(t-stid_bmap, t-nstids + t-nsftids); return 0; } @@ -2995,6 +3024,7 @@ static void uld_attach(struct adapter *adap, unsigned int uld) { void *handle; struct cxgb4_lld_info lli; + unsigned
[PATCH V3 5/5] RDMA/cxgb4: Fix bug for active and passive LE hash collision path
Retries active opens for INUSE errors. Logs any active ofld_connect_wr error replies. Sends ofld_connect_wr on same ctrlq. It needs to go on the same control txq as regular CPL active/passive messages. Retries on active open replies with EADDRINUSE. Uses active open fw wr only if active filter region is set. Adds stat for ofld_connect_wr failures. This patch also adds debugfs file to show endpoints. Signed-off-by: Vipul Pandya vi...@chelsio.com --- V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 drivers/infiniband/hw/cxgb4/cm.c| 277 --- drivers/infiniband/hw/cxgb4/device.c| 125 ++ drivers/infiniband/hw/cxgb4/iw_cxgb4.h | 32 +++ drivers/net/ethernet/chelsio/cxgb4/cxgb4.h |1 + drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c | 19 ++- drivers/net/ethernet/chelsio/cxgb4/cxgb4_uld.h |8 +- drivers/net/ethernet/chelsio/cxgb4/t4_regs.h|2 + 7 files changed, 376 insertions(+), 88 deletions(-) diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 2ac6a2c..b2ea840 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -275,6 +275,7 @@ void _c4iw_free_ep(struct kref *kref) cxgb4_remove_tid(ep-com.dev-rdev.lldi.tids, 0, ep-hwtid); dst_release(ep-dst); cxgb4_l2t_release(ep-l2t); + remove_handle(ep-com.dev, ep-com.dev-hwtid_idr, ep-hwtid); } kfree(ep); } @@ -550,6 +551,7 @@ static int send_connect(struct c4iw_ep *ep) req-opt0 = cpu_to_be64(opt0); req-params = cpu_to_be32(select_ntuple(ep-com.dev, ep-dst, ep-l2t)); req-opt2 = cpu_to_be32(opt2); + set_bit(ACT_OPEN_REQ, ep-com.history); return c4iw_l2t_send(ep-com.dev-rdev, skb, ep-l2t); } @@ -826,6 +828,7 @@ static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) /* setup the hwtid for this connection */ ep-hwtid = tid; cxgb4_insert_tid(t, ep, tid); + insert_handle(dev, dev-hwtid_idr, ep, ep-hwtid); ep-snd_seq = be32_to_cpu(req-snd_isn); ep-rcv_seq = be32_to_cpu(req-rcv_isn); @@ -833,7 +836,9 @@ static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb) set_emss(ep, ntohs(req-tcp_opt)); /* dealloc the atid */ + remove_handle(ep-com.dev, ep-com.dev-atid_idr, atid); cxgb4_free_atid(t, atid); + set_bit(ACT_ESTAB, ep-com.history); /* start MPA negotiation */ send_flowc(ep, NULL); @@ -859,6 +864,7 @@ static void close_complete_upcall(struct c4iw_ep *ep) ep-com.cm_id-rem_ref(ep-com.cm_id); ep-com.cm_id = NULL; ep-com.qp = NULL; + set_bit(CLOSE_UPCALL, ep-com.history); } } @@ -867,6 +873,7 @@ static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp) PDBG(%s ep %p tid %u\n, __func__, ep, ep-hwtid); close_complete_upcall(ep); state_set(ep-com, ABORTING); + set_bit(ABORT_CONN, ep-com.history); return send_abort(ep, skb, gfp); } @@ -881,6 +888,7 @@ static void peer_close_upcall(struct c4iw_ep *ep) PDBG(peer close delivered ep %p cm_id %p tid %u\n, ep, ep-com.cm_id, ep-hwtid); ep-com.cm_id-event_handler(ep-com.cm_id, event); + set_bit(DISCONN_UPCALL, ep-com.history); } } @@ -899,6 +907,7 @@ static void peer_abort_upcall(struct c4iw_ep *ep) ep-com.cm_id-rem_ref(ep-com.cm_id); ep-com.cm_id = NULL; ep-com.qp = NULL; + set_bit(ABORT_UPCALL, ep-com.history); } } @@ -931,6 +940,7 @@ static void connect_reply_upcall(struct c4iw_ep *ep, int status) PDBG(%s ep %p tid %u status %d\n, __func__, ep, ep-hwtid, status); + set_bit(CONN_RPL_UPCALL, ep-com.history); ep-com.cm_id-event_handler(ep-com.cm_id, event); if (status 0) { @@ -971,6 +981,7 @@ static void connect_request_upcall(struct c4iw_ep *ep) ep-parent_ep-com.cm_id, event); } + set_bit(CONNREQ_UPCALL, ep-com.history); c4iw_put_ep(ep-parent_ep-com); ep-parent_ep = NULL; } @@ -987,6 +998,7 @@ static void established_upcall(struct c4iw_ep *ep) if (ep-com.cm_id) { PDBG(%s ep %p tid %u\n, __func__, ep, ep-hwtid); ep-com.cm_id-event_handler(ep-com.cm_id, event); + set_bit(ESTAB_UPCALL, ep-com.history); } } @@ -1372,6 +1384,7 @@ static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb) unsigned int dlen = ntohs(hdr-len); unsigned int tid = GET_TID(hdr); struct tid_info *t =
[PATCH V3 3/5] RDMA/cxgb4: Fix LE hash collision bug for active open connection
It enables establishing active open connection using fw_ofld_connection work request when cpl_act_open_rpl says TCAM full error which may be because of LE hash collision. Current support is only for IPv4 active open connections. Sets ntuple bits in active open requests. For T4 firmware greater than 1.4.10.0 ntuple bits are required to be set. Adds nocong and enable_ecn module parameter options. Signed-off-by: Vipul Pandya vi...@chelsio.com --- V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 drivers/infiniband/hw/cxgb4/cm.c | 158 - drivers/infiniband/hw/cxgb4/iw_cxgb4.h|1 + drivers/net/ethernet/chelsio/cxgb4/t4_msg.h | 19 +++ drivers/net/ethernet/chelsio/cxgb4/t4_regs.h |2 + drivers/net/ethernet/chelsio/cxgb4/t4fw_api.h | 109 + 5 files changed, 283 insertions(+), 6 deletions(-) diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 5de8696..4878704 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -61,6 +61,14 @@ static char *states[] = { NULL, }; +static int nocong; +module_param(nocong, int, 0644); +MODULE_PARM_DESC(nocong, Turn of congestion control (default=0)); + +static int enable_ecn; +module_param(enable_ecn, int, 0644); +MODULE_PARM_DESC(enable_ecn, Enable ECN (default=0/disabled)); + static int dack_mode = 1; module_param(dack_mode, int, 0644); MODULE_PARM_DESC(dack_mode, Delayed ack mode (default=1)); @@ -441,6 +449,50 @@ static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp) return c4iw_l2t_send(ep-com.dev-rdev, skb, ep-l2t); } +#define VLAN_NONE 0xfff +#define FILTER_SEL_VLAN_NONE 0x +#define FILTER_SEL_WIDTH_P_FC (3+1) /* port uses 3 bits, FCoE one bit */ +#define FILTER_SEL_WIDTH_VIN_P_FC \ + (6 + 7 + FILTER_SEL_WIDTH_P_FC) /* 6 bits are unused, VF uses 7 bits*/ +#define FILTER_SEL_WIDTH_TAG_P_FC \ + (3 + FILTER_SEL_WIDTH_VIN_P_FC) /* PF uses 3 bits */ +#define FILTER_SEL_WIDTH_VLD_TAG_P_FC (1 + FILTER_SEL_WIDTH_TAG_P_FC) + +static unsigned int select_ntuple(struct c4iw_dev *dev, struct dst_entry *dst, + struct l2t_entry *l2t) +{ + unsigned int ntuple = 0; + u32 viid; + + switch (dev-rdev.lldi.filt_mode) { + + /* default filter mode */ + case HW_TPL_FR_MT_PR_IV_P_FC: + if (l2t-vlan == VLAN_NONE) + ntuple |= FILTER_SEL_VLAN_NONE FILTER_SEL_WIDTH_P_FC; + else { + ntuple |= l2t-vlan FILTER_SEL_WIDTH_P_FC; + ntuple |= 1 FILTER_SEL_WIDTH_VLD_TAG_P_FC; + } + ntuple |= l2t-lport S_PORT | IPPROTO_TCP + FILTER_SEL_WIDTH_VLD_TAG_P_FC; + break; + case HW_TPL_FR_MT_PR_OV_P_FC: { + viid = cxgb4_port_viid(l2t-neigh-dev); + + ntuple |= FW_VIID_VIN_GET(viid) FILTER_SEL_WIDTH_P_FC; + ntuple |= FW_VIID_PFN_GET(viid) FILTER_SEL_WIDTH_VIN_P_FC; + ntuple |= FW_VIID_VIVLD_GET(viid) FILTER_SEL_WIDTH_TAG_P_FC; + ntuple |= l2t-lport S_PORT | IPPROTO_TCP + FILTER_SEL_WIDTH_VLD_TAG_P_FC; + break; + } + default: + break; + } + return ntuple; +} + static int send_connect(struct c4iw_ep *ep) { struct cpl_act_open_req *req; @@ -463,7 +515,8 @@ static int send_connect(struct c4iw_ep *ep) cxgb4_best_mtu(ep-com.dev-rdev.lldi.mtus, ep-mtu, mtu_idx); wscale = compute_wscale(rcv_win); - opt0 = KEEP_ALIVE(1) | + opt0 = (nocong ? NO_CONG(1) : 0) | + KEEP_ALIVE(1) | DELACK(1) | WND_SCALE(wscale) | MSS_IDX(mtu_idx) | @@ -474,6 +527,7 @@ static int send_connect(struct c4iw_ep *ep) ULP_MODE(ULP_MODE_TCPDDP) | RCV_BUFSIZ(rcv_win10); opt2 = RX_CHANNEL(0) | + CCTRL_ECN(enable_ecn) | RSS_QUEUE_VALID | RSS_QUEUE(ep-rss_qid); if (enable_tcp_timestamps) opt2 |= TSTAMPS_EN(1); @@ -492,7 +546,7 @@ static int send_connect(struct c4iw_ep *ep) req-local_ip = ep-com.local_addr.sin_addr.s_addr; req-peer_ip = ep-com.remote_addr.sin_addr.s_addr; req-opt0 = cpu_to_be64(opt0); - req-params = 0; + req-params = cpu_to_be32(select_ntuple(ep-com.dev, ep-dst, ep-l2t)); req-opt2 = cpu_to_be32(opt2); return c4iw_l2t_send(ep-com.dev-rdev, skb, ep-l2t); } @@ -1383,6 +1437,61 @@ static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb) return 0; } +static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid) +{ + struct sk_buff *skb; + struct fw_ofld_connection_wr *req; + unsigned int mtu_idx; +
[PATCH V3 4/5] RDMA/cxgb4: Fix LE hash collision bug for passive open connection
It establishes passive open connection through firmware work request. Passive open connection will go through this path as now instead of listening server we create a server filter which will redirect the incoming SYN packet to the offload queue. After this driver tries to establish the connection using firmware work request. Signed-off-by: Vipul Pandya vi...@chelsio.com --- V2: Changed commenting style from kernel-doc format('/**') to normal V3: Resending the whole patch series again with the missing patches in V2 drivers/infiniband/hw/cxgb4/cm.c| 411 --- drivers/infiniband/hw/cxgb4/device.c| 85 +- drivers/net/ethernet/chelsio/cxgb4/cxgb4_main.c |9 + drivers/net/ethernet/chelsio/cxgb4/t4_msg.h | 46 +++ 4 files changed, 496 insertions(+), 55 deletions(-) diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c index 4878704..2ac6a2c 100644 --- a/drivers/infiniband/hw/cxgb4/cm.c +++ b/drivers/infiniband/hw/cxgb4/cm.c @@ -38,10 +38,12 @@ #include linux/inetdevice.h #include linux/ip.h #include linux/tcp.h +#include linux/if_vlan.h #include net/neighbour.h #include net/netevent.h #include net/route.h +#include net/tcp.h #include iw_cxgb4.h @@ -1569,13 +1571,14 @@ static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb) struct c4iw_listen_ep *ep = lookup_stid(t, stid); if (!ep) { - printk(KERN_ERR MOD stid %d lookup failure!\n, stid); - return 0; + PDBG(%s stid %d lookup failure!\n, __func__, stid); + goto out; } PDBG(%s ep %p status %d error %d\n, __func__, ep, rpl-status, status2errno(rpl-status)); c4iw_wake_up(ep-com.wr_wait, status2errno(rpl-status)); +out: return 0; } @@ -1779,15 +1782,23 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) unsigned int hwtid = GET_TID(req); struct dst_entry *dst; struct rtable *rt; - __be32 local_ip, peer_ip; + __be32 local_ip, peer_ip = 0; __be16 local_port, peer_port; int err; + u16 peer_mss = ntohs(req-tcpopt.mss); parent_ep = lookup_stid(t, stid); - PDBG(%s parent ep %p tid %u\n, __func__, parent_ep, hwtid); - + if (!parent_ep) { + PDBG(%s connect request on invalid stid %d\n, __func__, stid); + goto reject; + } get_4tuple(req, local_ip, peer_ip, local_port, peer_port); + PDBG(%s parent ep %p hwtid %u laddr 0x%x raddr 0x%x lport %d \ +rport %d peer_mss %d\n, __func__, parent_ep, hwtid, +ntohl(local_ip), ntohl(peer_ip), ntohs(local_port), +ntohs(peer_port), peer_mss); + if (state_read(parent_ep-com) != LISTEN) { printk(KERN_ERR %s - listening ep not in LISTEN\n, __func__); @@ -1821,6 +1832,9 @@ static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb) goto reject; } + if (peer_mss child_ep-mtu (peer_mss + 40)) + child_ep-mtu = peer_mss + 40; + state_set(child_ep-com, CONNECTING); child_ep-com.dev = dev; child_ep-com.cm_id = NULL; @@ -1861,6 +1875,9 @@ static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb) ep-snd_seq = be32_to_cpu(req-snd_isn); ep-rcv_seq = be32_to_cpu(req-rcv_isn); + PDBG(%s ep %p hwtid %u tcp_opt 0x%02x\n, __func__, ep, tid, +ntohs(req-tcp_opt)); + set_emss(ep, ntohs(req-tcp_opt)); dst_confirm(ep-dst); @@ -2478,7 +2495,6 @@ int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) struct c4iw_dev *dev = to_c4iw_dev(cm_id-device); struct c4iw_listen_ep *ep; - might_sleep(); ep = alloc_ep(sizeof(*ep), GFP_KERNEL); @@ -2497,30 +2513,49 @@ int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog) /* * Allocate a server TID. */ - ep-stid = cxgb4_alloc_stid(dev-rdev.lldi.tids, PF_INET, ep); + if (dev-rdev.lldi.enable_fw_ofld_conn) + ep-stid = cxgb4_alloc_sftid(dev-rdev.lldi.tids, PF_INET, ep); + else + ep-stid = cxgb4_alloc_stid(dev-rdev.lldi.tids, PF_INET, ep); + if (ep-stid == -1) { printk(KERN_ERR MOD %s - cannot alloc stid.\n, __func__); err = -ENOMEM; goto fail2; } - state_set(ep-com, LISTEN); - c4iw_init_wr_wait(ep-com.wr_wait); - err = cxgb4_create_server(ep-com.dev-rdev.lldi.ports[0], ep-stid, - ep-com.local_addr.sin_addr.s_addr, - ep-com.local_addr.sin_port, - ep-com.dev-rdev.lldi.rxq_ids[0]); - if (err) - goto fail3; - - /* wait for pass_open_rpl */ - err = c4iw_wait_for_reply(ep-com.dev-rdev,
Re: Dapltest test error DAT_CONN_QUAL_IN_USE
Hi Arlin, There was already a bug logged in openfabrics bugzilla regarding this. Following is a link for the same. http://bugs.openfabrics.org/bugzilla/show_bug.cgi?id=2400 I have assigned this bug to your name. Thanks, Vipul On 01-12-2012 01:46, Davis, Arlin R wrote: http://openfabrics.org/bugzilla/index.cgi -Original Message- From: linux-rdma-ow...@vger.kernel.org [mailto:linux-rdma- ow...@vger.kernel.org] On Behalf Of Vipul Pandya Sent: Friday, November 30, 2012 7:12 AM To: Davis, Arlin R Cc: Steve Wise; linux-rdma@vger.kernel.org; Kumar A S; Abhishek Agrawal; Divy Le Ray Subject: Re: Dapltest test error DAT_CONN_QUAL_IN_USE Arlin, Can you please refer to which bugzilla I should log a bug? Can you please provide me the url? Thanks, Vipul On 30-11-2012 05:21, Davis, Arlin R wrote: Vipul, Can you submit a bug in bugzilla for tracking? I will try to get to this next couple of days. -arlin -Original Message- From: Vipul Pandya [mailto:vi...@chelsio.com] Sent: Thursday, November 29, 2012 5:34 AM To: Davis, Arlin R Cc: Steve Wise; linux-rdma@vger.kernel.org; Kumar A S; Abhishek Agrawal; Divy Le Ray Subject: Re: Dapltest test error DAT_CONN_QUAL_IN_USE Hi Arlin, This issue is happening because there is a port collision between dapltest server port space and host TCP stack. The port collision happens because rdma_bind_addr is getting called from the two different places with different port arguments from dapltest. rdma_bind_addr is getting called from the following two places: 1. Once it is getting called from dapls_ib_setup_conn_listener function with starting port as 45278. Based on number of threads and eps, in subsequent call of dapls_ib_setup_conn_listener this port number will keep getting incremented. 2. 2nd time it is getting called from dapls_ib_qp_alloc function with port number as always 0. Now, when rdma_bind_addr gets called with port number 0 it will allocate any free random port number. Then when dapls_ib_setup_conn_listener calls the rdma_bind_addr with fix port number which is already allocate via dapls_ib_qp_alloc function rdma_bind_addr will return EADDRINUSE error, which in turn will result in DAT_CONN_QUAL_IN_USE error. I think solution here would be to call rdma_bind_addr from both the location passing port number from the same port range. Please let me know your thoughts on this. Our testing has been blocked because of this issue. We would like to get this fixed. Please let us know if we need to log a bug anywhere for this. Thanks, Vipul On 27-11-2012 01:24, Steve Wise wrote: Perhaps the port is in use by the host TCP stack? On 11/26/2012 1:30 PM, Davis, Arlin R wrote: dapltest server will start with port 45278 and increase by client thread count during each new client connection. If you never restart the server it will continue to increase the listen port based on new clients connecting. If you restart dapltest it will restart back at port 45278. I am not familiar with iWarp CM but the error is coming from rdma_bind_addr (EADDRINUSE|EBUSY|EADDRNOTAVAIL). I will have to defer to Steve for this error. -arlin -Original Message- From: linux-rdma-ow...@vger.kernel.org [mailto:linux-rdma- ow...@vger.kernel.org] On Behalf Of Vipul Pandya Sent: Friday, November 23, 2012 5:54 AM To: linux-rdma@vger.kernel.org Cc: Kumar A S; Steve Wise; Abhishek Agrawal; Davis, Arlin R; Divy Le Ray Subject: Dapltest test error DAT_CONN_QUAL_IN_USE Hi All, I was running dapltest between my client and server machines with OFED- 3.5. While running the test it dapltest server throws an error DAT_CONN_QUAL_IN_USE if I increase number of threads and endpoints. Dapltest server: --- dapltest -T S -D chelsio1 Dapltest client: --- dapltest -T T -s 102.1.1.2 -D chelsio1 -R BE -i 1 -t 16 -w 8 server SR 8192 4 client SR 8192 4 Once I run the above test i get the following error on server side and client side stalls. $# dapltest -T S -D chelsio1 Dapltest: Service Point Ready - chelsio1 Test[b13f]: dat_psp_create #6 error: DAT_CONN_QUAL_IN_USE Test[b13f]: Warning: dat_ep_disconnect (abrupt) #0 error DAT_INVALID_STATE DAT_INVALID_STATE_EP_UNCONNECTED Test[b13f]: dat_evd_free (creq) error: DAT_INVALID_STATE DAT_INVALID_STATE_EVD_IN_USE Test[b13f]: Warning: dat_ep_disconnect (abrupt) #1 error DAT_INVALID_STATE DAT_INVALID_STATE_EP_UNCONNECTED Test[b13f]: dat_evd_free (creq) error: DAT_INVALID_STATE DAT_INVALID_STATE_EVD_IN_USE Test[b13f]: Warning: dat_ep_disconnect (abrupt) #2 error DAT_INVALID_STATE DAT_INVALID_STATE_EP_UNCONNECTED Test[b13f]: dat_evd_free (creq) error: DAT_INVALID_STATE DAT_INVALID_STATE_EVD_IN_USE Test[b13f]: Warning: dat_ep_disconnect (abrupt) #3 error DAT_INVALID_STATE DAT_INVALID_STATE_EP_UNCONNECTED Test[b13f]: dat_evd_free (creq) error: DAT_INVALID_STATE DAT_INVALID_STATE_EVD_IN_USE Test[b13f]: Warning:
[PATCH] opensm/osm_req.c: In req_determine_mkey, add more info when ERR 1107 occurs
Also dump DR path of SM class query Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_req.c |1 + 1 files changed, 1 insertions(+), 0 deletions(-) diff --git a/opensm/osm_req.c b/opensm/osm_req.c index 1c3c172..1f5db3f 100644 --- a/opensm/osm_req.c +++ b/opensm/osm_req.c @@ -106,6 +106,7 @@ static ib_net64_t req_determine_mkey(IN osm_sm_t * sm, OSM_LOG(sm-p_log, OSM_LOG_ERROR, req_determine_mkey: ERR 1107: Outgoing physp is null on non-hop_0!\n); + osm_dump_dr_path_v2(sm-p_log, p_path, FILE_ID, OSM_LOG_ERROR); dest_port_guid = 0; goto Remote_Guid; } -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH 1/2] opensm/osm_req.c: In req_determine_mkey, fix DR algorithm
so that it works regardless of SM port type Now works on switch port 0 as well as CA and router ports Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_req.c | 12 1 files changed, 8 insertions(+), 4 deletions(-) diff --git a/opensm/osm_req.c b/opensm/osm_req.c index 1f5db3f..b48be48 100644 --- a/opensm/osm_req.c +++ b/opensm/osm_req.c @@ -87,16 +87,20 @@ static ib_net64_t req_determine_mkey(IN osm_sm_t * sm, goto Remote_Guid; } - if (p_sm_port) + if (p_sm_port) { + /* get the node for the SM */ + p_node = p_sm_port-p_node; p_physp = p_sm_port-p_physp; + } - /* hop_count == 1: outgoing physp is SM physp */ - for (hop = 2; p_physp hop = p_path-hop_count; hop++) { + for (hop = 1; p_physp hop = p_path-hop_count; hop++) { + p_physp = osm_node_get_physp_ptr(p_node, p_path-path[hop]); + if (!p_physp) + break; p_physp = p_physp-p_remote_physp; if (!p_physp) break; p_node = p_physp-p_node; - p_physp = osm_node_get_physp_ptr(p_node, p_path-path[hop]); } /* At this point, p_physp points at the outgoing physp on the -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH 2/2] opensm/osm_req.c: In req_determine_mkey, use osm_physp APIs
rather tha direct accesses to physp struct Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_req.c |4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-) diff --git a/opensm/osm_req.c b/opensm/osm_req.c index b48be48..caf94fb 100644 --- a/opensm/osm_req.c +++ b/opensm/osm_req.c @@ -97,10 +97,10 @@ static ib_net64_t req_determine_mkey(IN osm_sm_t * sm, p_physp = osm_node_get_physp_ptr(p_node, p_path-path[hop]); if (!p_physp) break; - p_physp = p_physp-p_remote_physp; + p_physp = osm_physp_get_remote(p_physp); if (!p_physp) break; - p_node = p_physp-p_node; + p_node = osm_physp_get_node_ptr(p_physp); } /* At this point, p_physp points at the outgoing physp on the -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] opensm/osm_subnet.c: Improve error messages in subn_validate_neighbor
Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_subnet.c | 22 -- 1 files changed, 16 insertions(+), 6 deletions(-) diff --git a/opensm/osm_subnet.c b/opensm/osm_subnet.c index a3fc0b5..c716839 100644 --- a/opensm/osm_subnet.c +++ b/opensm/osm_subnet.c @@ -946,7 +946,8 @@ static void subn_validate_neighbor(osm_subn_t *p_subn) valid_entry = TRUE; OSM_LOG((p_subn-p_osm-log), OSM_LOG_DEBUG, - Validating neighbor for 0x%016 PRIx64 , port %d\n, + Validating neighbor for guid:0x%016 PRIx64 + , port %d\n, p_item-guid, p_item-portnum); if (p_item-guid == 0) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, @@ -954,31 +955,40 @@ static void subn_validate_neighbor(osm_subn_t *p_subn) valid_entry = FALSE; } else if (p_item-portnum == 0) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, - ERR 7514: found invalid zero port\n); + ERR 7514: found invalid zero port for + guid: 0x%016 PRIx64 \n, + p_item-guid); valid_entry = FALSE; } else if (osm_db_neighbor_get(p_subn-p_neighbor, p_item-guid, p_item-portnum, guid, port)) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, ERR 7515: could not find neighbor for - guid: 0x%016 PRIx64 \n, p_item-guid); + guid: 0x%016 PRIx64 , port %d\n, + p_item-guid, p_item-portnum); valid_entry = FALSE; } else if (guid == 0) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, ERR 7516: found invalid neighbor - zero guid); + zero guid for guid: 0x%016 PRIx64 + , port %d\n, + p_item-guid, p_item-portnum); valid_entry = FALSE; } else if (port == 0) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, ERR 7517: found invalid neighbor - zero port\n); + zero port for guid: 0x%016 PRIx64 + , port %d\n, + p_item-guid, p_item-portnum); valid_entry = FALSE; } else if (osm_db_neighbor_get(p_subn-p_neighbor, guid, port, guid, port) || guid != p_item-guid || port != p_item-portnum) { OSM_LOG((p_subn-p_osm-log), OSM_LOG_ERROR, ERR 7518: neighbor does not point - back at us\n); + back at us (guid: 0x%016 PRIx64 + , port %d)\n, + p_item-guid, p_item-portnum); valid_entry = FALSE; } -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] opensm/osm_ucast_ftree.c: Remove duplicate free in fabric_create_leaf_switch_array
Signed-off-by: Dan Ben-Yosef da...@mellanox.com Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_ucast_ftree.c |1 - 1 files changed, 0 insertions(+), 1 deletions(-) diff --git a/opensm/osm_ucast_ftree.c b/opensm/osm_ucast_ftree.c index 532a29f..133fd4f 100644 --- a/opensm/osm_ucast_ftree.c +++ b/opensm/osm_ucast_ftree.c @@ -1689,7 +1689,6 @@ static int fabric_create_leaf_switch_array(IN ftree_fabric_t * p_ftree) if (!p_ftree-leaf_switches) { osm_log_v2(p_ftree-p_osm-log, OSM_LOG_SYS, FILE_ID, Fat-tree routing: Memory allocation failed\n); - free(all_switches_at_leaf_level); res = -1; goto Exit; } -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] opensm/osm_ucast_ftree.c: Eliminate unneeded NULL pointer checks prior to calls to free
Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_ucast_ftree.c | 20 ++-- 1 files changed, 6 insertions(+), 14 deletions(-) diff --git a/opensm/osm_ucast_ftree.c b/opensm/osm_ucast_ftree.c index 133fd4f..d58fff1 100644 --- a/opensm/osm_ucast_ftree.c +++ b/opensm/osm_ucast_ftree.c @@ -341,8 +341,6 @@ static ftree_sw_tbl_element_t *sw_tbl_element_create(IN ftree_sw_t * p_sw) static void sw_tbl_element_destroy(IN ftree_sw_tbl_element_t * p_element) { - if (!p_element) - return; free(p_element); } @@ -370,8 +368,7 @@ static ftree_port_t *port_create(IN uint8_t port_num, static void port_destroy(IN ftree_port_t * p_port) { - if (p_port) - free(p_port); + free(p_port); } /*** @@ -609,12 +606,9 @@ static void sw_destroy(IN ftree_fabric_t * p_ftree, IN ftree_sw_t * p_sw) port_group_destroy(p_sw-sibling_port_groups[i]); for (i = 0; i p_sw-up_port_groups_num; i++) port_group_destroy(p_sw-up_port_groups[i]); - if (p_sw-down_port_groups) - free(p_sw-down_port_groups); - if (p_sw-sibling_port_groups) - free(p_sw-sibling_port_groups); - if (p_sw-up_port_groups) - free(p_sw-up_port_groups); + free(p_sw-down_port_groups); + free(p_sw-sibling_port_groups); + free(p_sw-up_port_groups); free(p_sw); } /* sw_destroy() */ @@ -829,8 +823,7 @@ static void hca_destroy(IN ftree_hca_t * p_hca) for (i = 0; i p_hca-up_port_groups_num; i++) port_group_destroy(p_hca-up_port_groups[i]); - if (p_hca-up_port_groups) - free(p_hca-up_port_groups); + free(p_hca-up_port_groups); free(p_hca); } @@ -1702,8 +1695,7 @@ static int fabric_create_leaf_switch_array(IN ftree_fabric_t * p_ftree) p_ftree-leaf_switches_num); Exit: - if (all_switches_at_leaf_level) - free(all_switches_at_leaf_level); + free(all_switches_at_leaf_level); OSM_LOG_EXIT(p_ftree-p_osm-log); return res; } /* fabric_create_leaf_switch_array() */ -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
Signed-off-by: Alex Netes ale...@mellanox.com Signed-off-by: Hal Rosenstock h...@mellanox.com --- opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ - if (!sw || !port_num || + if (!sw || !port_num || sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH V3 1/5] cxgb4: Add T4 filter support
From: Vipul Pandya vi...@chelsio.com Date: Mon, 3 Dec 2012 16:52:57 +0530 + /* + * If the new or old filter have loopback rewriteing rules then we'll + * need to free any existing Layer Two Table (L2T) entries of the old + * filter rule. The firmware will handle freeing up any Source MAC + * Table (SMT) entries used for rewriting Source MAC Addresses in + * loopback rules. + */ + if (f-l2t) + cxgb4_l2t_release(f-l2t); + + /* The zeroing of the filter rule below clears the filter valid, + * pending, locked flags, l2t pointer, etc. so it's all we need for + * this operation. + */ My greatest amusement is achieved when someone gets a code formatting rule both correct and wrong in the same patch, as you have done here. Format comments in the networking: /* Like * this. */ /* * Never, * like this. */ Thanks. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: [PATCH librdmacm] Fixed build problem due to missing macro
Thanks - applied. I'm using a locally modified version of verbs, so I don't end up seeing this error. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH librdmacm] Fixed build problem due to missing macro
On Mon, Dec 3, 2012 at 9:35 PM, Hefty, Sean sean.he...@intel.com wrote: I'm using a locally modified version of verbs, so I don't end up seeing this error. could this be related to different results while running netperf .w fork in our setup vs. yours? Or. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: [PATCH librdmacm] Fixed build problem due to missing macro
I'm using a locally modified version of verbs, so I don't end up seeing this error. could this be related to different results while running netperf .w fork in our setup vs. yours? I was referring to the build issue, not the fork problem. According to my notes, I ran netperf with fork support using this: server -- export RDMAV_FORK_SAFE=1 LD_PRELOAD=/usr/local/lib/rsocket/librspreload.so netserver client -- export RDMAV_FORK_SAFE=1 LD_PRELOAD=/usr/local/lib/rsocket/librspreload.so netperf -H ip-address -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] infiniband-diags: Remove Red Hat-ism.
Future release of Fedora are going to remove /etc/sysconfig/network which we source to get $HOSTNAME. Bash (and sh) set $HOSTNAME in the shell by default, so we should be safe to use that here. This adds the benefit of working across multiple distributions if that is required in the future. Related: rhbz881917 Signed-off-by: Jon Stanley jonstan...@gmail.com Cc: Doug Ledford dledf...@redhat.com Cc: Ira Weiny wei...@llnl.gov --- scripts/set_nodedesc.sh |3 --- 1 files changed, 0 insertions(+), 3 deletions(-) diff --git a/scripts/set_nodedesc.sh b/scripts/set_nodedesc.sh index 1e42ac8..3d2e406 100755 --- a/scripts/set_nodedesc.sh +++ b/scripts/set_nodedesc.sh @@ -1,8 +1,5 @@ #!/bin/sh -if [ -f /etc/sysconfig/network ]; then -. /etc/sysconfig/network -fi ib_sysfs=/sys/class/infiniband newname=$HOSTNAME -- 1.7.1 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
On 12/3/2012 4:09 PM, Jim Schutt wrote: On 12/03/2012 08:26 AM, Hal Rosenstock wrote: Signed-off-by: Alex Netesale...@mellanox.com Signed-off-by: Hal Rosenstockh...@mellanox.com --- opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ -if (!sw || !port_num || +if (!sw || !port_num || sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; With the patch as-is, if torus_update_osm_vlarb() returns early for any non-NULL switch port, it will never do any updates. If the crash was that sw-port[port_num] was NULL, shouldn't the check be !sw-port[port_num] ? Can you tell me more about the test case that leads to the crash? Is it that there's a switch with a port that's not connected to anything, and torus_update_osm_vlarb() was called for it? Testing for a non-NULL sw-port[port_num] is definitely the right thing to do to handle that case, and I'm sorry I missed it earlier. If not, then something else is likely broken, and we need to find and fix that. Yes, it was meant to be NULL pointer check. v2 of patch coming soon. -- Hal -- Jim -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
Signed-off-by: Alex Netes ale...@mellanox.com Signed-off-by: Hal Rosenstock h...@mellanox.com --- Change since v1: Fixed NULL pointer check on sw-port[port_num] Pointed out by: Jim Schutt jasc...@sandia.gov opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ - if (!sw || !port_num || + if (!sw || !port_num || !sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCHv2} opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
Signed-off-by: Alex Netes ale...@mellanox.com Signed-off-by: Hal Rosenstock h...@mellanox.com --- Change since v1: Fixed NULL pointer check on sw-port[port_num] Pointed out by: Jim Schutt jasc...@sandia.gov opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ - if (!sw || !port_num || + if (!sw || !port_num || !sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; -- 1.7.8.2 -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH] opensm/osm_port_info_rcv.c: Don't modify subnet minimal values when PortState DOWN
as they are vendor dependent fields in PortInfo and can't be relied on due to: C14-24.2.1: If PortInfo:Portstate=Down, then o a SubnGet(PortInfo) shall produce valid data for PortInfo:PortState and PortInfo:PortPhysicalState; whether any other component has valid data is vendor-dependent. Signed-off-by: Hal Rosenstock h...@mellanox.com --- diff --git a/opensm/osm_port_info_rcv.c b/opensm/osm_port_info_rcv.c index 6f6922c..859a6be 100644 --- a/opensm/osm_port_info_rcv.c +++ b/opensm/osm_port_info_rcv.c @@ -94,7 +94,8 @@ static void pi_rcv_process_endport(IN osm_sm_t * sm, IN osm_physp_t * p_physp, port_guid = osm_physp_get_port_guid(p_physp); /* HACK extended port 0 should be handled too! */ - if (osm_physp_get_port_num(p_physp) != 0) { + if (osm_physp_get_port_num(p_physp) != 0 + ib_port_info_get_port_state(p_pi) != IB_LINK_DOWN) { /* track the minimal endport MTU, rate, and operational VLs */ mtu = ib_port_info_get_mtu_cap(p_pi); if (mtu sm-p_subn-min_ca_mtu) { @@ -323,6 +324,9 @@ static void pi_rcv_process_switch_port(IN osm_sm_t * sm, IN osm_node_t * p_node, p_physp-port_info = *p_pi; pi_rcv_process_endport(sm, p_physp, p_pi); } else { + if (ib_port_info_get_port_state(p_pi) == IB_LINK_DOWN) + goto Exit; + /* track the minimal switch ext port MTU, rate, and operational VLs */ mtu = ib_port_info_get_mtu_cap(p_pi); if (mtu sm-p_subn-min_sw_mtu) { @@ -358,6 +362,7 @@ static void pi_rcv_process_switch_port(IN osm_sm_t * sm, IN osm_node_t * p_node, } } +Exit: OSM_LOG_EXIT(sm-p_log); } -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
On 12/03/2012 08:26 AM, Hal Rosenstock wrote: Signed-off-by: Alex Netesale...@mellanox.com Signed-off-by: Hal Rosenstockh...@mellanox.com --- opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ - if (!sw || !port_num || + if (!sw || !port_num || sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; With the patch as-is, if torus_update_osm_vlarb() returns early for any non-NULL switch port, it will never do any updates. If the crash was that sw-port[port_num] was NULL, shouldn't the check be !sw-port[port_num] ? Can you tell me more about the test case that leads to the crash? Is it that there's a switch with a port that's not connected to anything, and torus_update_osm_vlarb() was called for it? Testing for a non-NULL sw-port[port_num] is definitely the right thing to do to handle that case, and I'm sorry I missed it earlier. If not, then something else is likely broken, and we need to find and fix that. -- Jim -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCHv2} opensm/osm_torus.c: Fix crash in torus_update_osm_vlarb
On 12/03/2012 02:25 PM, Hal Rosenstock wrote: Signed-off-by: Alex Netesale...@mellanox.com Signed-off-by: Hal Rosenstockh...@mellanox.com Acked-by: Jim Schutt jasc...@sandia.gov --- Change since v1: Fixed NULL pointer check on sw-port[port_num] Pointed out by: Jim Schuttjasc...@sandia.gov opensm/osm_torus.c |2 +- 1 files changed, 1 insertions(+), 1 deletions(-) diff --git a/opensm/osm_torus.c b/opensm/osm_torus.c index c06f8d4..075f84a 100644 --- a/opensm/osm_torus.c +++ b/opensm/osm_torus.c @@ -8089,7 +8089,7 @@ void torus_update_osm_vlarb(void *context, osm_physp_t *osm_phys_port, * So, leave VL 0 alone, remap VL 4 to VL 1, zero out the rest, * and compress out the zero entries to the end. */ - if (!sw || !port_num || + if (!sw || !port_num || !sw-port[port_num] || sw-port[port_num]-pgrp-port_grp != 2 * TORUS_MAX_DIM) return; -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH] infiniband-diags: Remove Red Hat-ism.
On Mon, Dec 03, 2012 at 03:28:18PM -0500, Jon Stanley wrote: Future release of Fedora are going to remove /etc/sysconfig/network which we source to get $HOSTNAME. Bash (and sh) set $HOSTNAME in the shell by default, so we should be safe to use that here. This adds the benefit of working across multiple distributions if that is required in the future. This looks like it is a bashism (ie dash does not do this), so you should update the #! to use /bin/bash not /bin/sh... $ HOSTNAME= /bin/dash \h{\u}\w#echo $HOSTNAME \h{\u}\w# Jason -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html