diff options
author | Michael J. Ruhl <michael.j.ruhl@intel.com> | 2017-07-24 07:46:12 -0700 |
---|---|---|
committer | Doug Ledford <dledford@redhat.com> | 2017-07-31 15:17:55 -0400 |
commit | bf90aadd630c2c9f7f965ba1e90d41b5b46db7c9 (patch) | |
tree | e95f69b0a661f5693efe4a4d1700f4243f1d6983 | |
parent | 2250563e2c935d6401a2203be4de4ca2cf0db183 (diff) | |
download | linux-bf90aadd630c2c9f7f965ba1e90d41b5b46db7c9.tar.bz2 |
IB/hfi1: Send MAD traps until repressed
A trap should be sent to the FM until the FM sends a repress message.
This is in line with the IBTA 13.4.9.
Add the ability to resend traps until a repress message is received.
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Reviewed-by: Michael N. Henry <michael.n.henry@intel.com>
Signed-off-by: Michael J. Ruhl <michael.j.ruhl@intel.com>
Signed-off-by: Dennis Dalessandro <dennis.dalessandro@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
-rw-r--r-- | drivers/infiniband/hw/hfi1/mad.c | 373 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/mad.h | 3 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/verbs.c | 5 | ||||
-rw-r--r-- | include/rdma/rdma_vt.h | 17 |
4 files changed, 310 insertions, 88 deletions
diff --git a/drivers/infiniband/hw/hfi1/mad.c b/drivers/infiniband/hw/hfi1/mad.c index a081a98d728a..0a3e2dfdf56e 100644 --- a/drivers/infiniband/hw/hfi1/mad.c +++ b/drivers/infiniband/hw/hfi1/mad.c @@ -59,6 +59,16 @@ #define OPA_LINK_WIDTH_RESET_OLD 0x0fff #define OPA_LINK_WIDTH_RESET 0xffff +struct trap_node { + struct list_head list; + struct opa_mad_notice_attr data; + __be64 tid; + int len; + u32 retry; + u8 in_use; + u8 repress; +}; + static int smp_length_check(u32 data_size, u32 request_len) { if (unlikely(request_len < data_size)) @@ -97,28 +107,156 @@ void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port) ib_dispatch_event(&event); } -static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len) +/* + * If the port is down, clean up all pending traps. We need to be careful + * with the given trap, because it may be queued. + */ +static void cleanup_traps(struct hfi1_ibport *ibp, struct trap_node *trap) +{ + struct trap_node *node, *q; + unsigned long flags; + struct list_head trap_list; + int i; + + for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) { + spin_lock_irqsave(&ibp->rvp.lock, flags); + list_replace_init(&ibp->rvp.trap_lists[i].list, &trap_list); + ibp->rvp.trap_lists[i].list_len = 0; + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + + /* + * Remove all items from the list, freeing all the non-given + * traps. + */ + list_for_each_entry_safe(node, q, &trap_list, list) { + list_del(&node->list); + if (node != trap) + kfree(node); + } + } + + /* + * If this wasn't on one of the lists it would not be freed. If it + * was on the list, it is now safe to free. + */ + kfree(trap); +} + +static struct trap_node *check_and_add_trap(struct hfi1_ibport *ibp, + struct trap_node *trap) +{ + struct trap_node *node; + struct trap_list *trap_list; + unsigned long flags; + unsigned long timeout; + int found = 0; + + /* + * Since the retry (handle timeout) does not remove a trap request + * from the list, all we have to do is compare the node. + */ + spin_lock_irqsave(&ibp->rvp.lock, flags); + trap_list = &ibp->rvp.trap_lists[trap->data.generic_type & 0x0F]; + + list_for_each_entry(node, &trap_list->list, list) { + if (node == trap) { + node->retry++; + found = 1; + break; + } + } + + /* If it is not on the list, add it, limited to RVT-MAX_TRAP_LEN. */ + if (!found) { + if (trap_list->list_len < RVT_MAX_TRAP_LEN) { + trap_list->list_len++; + list_add_tail(&trap->list, &trap_list->list); + } else { + pr_warn_ratelimited("hfi1: Maximim trap limit reached for 0x%0x traps\n", + trap->data.generic_type); + kfree(trap); + } + } + + /* + * Next check to see if there is a timer pending. If not, set it up + * and get the first trap from the list. + */ + node = NULL; + if (!timer_pending(&ibp->rvp.trap_timer)) { + /* + * o14-2 + * If the time out is set we have to wait until it expires + * before the trap can be sent. + * This should be > RVT_TRAP_TIMEOUT + */ + timeout = (RVT_TRAP_TIMEOUT * + (1UL << ibp->rvp.subnet_timeout)) / 1000; + mod_timer(&ibp->rvp.trap_timer, + jiffies + usecs_to_jiffies(timeout)); + node = list_first_entry(&trap_list->list, struct trap_node, + list); + node->in_use = 1; + } + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + + return node; +} + +static void subn_handle_opa_trap_repress(struct hfi1_ibport *ibp, + struct opa_smp *smp) +{ + struct trap_list *trap_list; + struct trap_node *trap; + unsigned long flags; + int i; + + if (smp->attr_id != IB_SMP_ATTR_NOTICE) + return; + + spin_lock_irqsave(&ibp->rvp.lock, flags); + for (i = 0; i < RVT_MAX_TRAP_LISTS; i++) { + trap_list = &ibp->rvp.trap_lists[i]; + trap = list_first_entry_or_null(&trap_list->list, + struct trap_node, list); + if (trap && trap->tid == smp->tid) { + if (trap->in_use) { + trap->repress = 1; + } else { + trap_list->list_len--; + list_del(&trap->list); + kfree(trap); + } + break; + } + } + spin_unlock_irqrestore(&ibp->rvp.lock, flags); +} + +static void send_trap(struct hfi1_ibport *ibp, struct trap_node *trap) { struct ib_mad_send_buf *send_buf; struct ib_mad_agent *agent; struct opa_smp *smp; - int ret; unsigned long flags; - unsigned long timeout; int pkey_idx; u32 qpn = ppd_from_ibp(ibp)->sm_trap_qp; agent = ibp->rvp.send_agent; - if (!agent) + if (!agent) { + cleanup_traps(ibp, trap); return; + } /* o14-3.2.1 */ - if (driver_lstate(ppd_from_ibp(ibp)) != IB_PORT_ACTIVE) + if (driver_lstate(ppd_from_ibp(ibp)) != IB_PORT_ACTIVE) { + cleanup_traps(ibp, trap); return; + } - /* o14-2 */ - if (ibp->rvp.trap_timeout && time_before(jiffies, - ibp->rvp.trap_timeout)) + /* Add the trap to the list if necessary and see if we can send it */ + trap = check_and_add_trap(ibp, trap); + if (!trap) return; pkey_idx = hfi1_lookup_pkey_idx(ibp, LIM_MGMT_P_KEY); @@ -139,11 +277,21 @@ static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len) smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED; smp->class_version = OPA_SM_CLASS_VERSION; smp->method = IB_MGMT_METHOD_TRAP; - ibp->rvp.tid++; - smp->tid = cpu_to_be64(ibp->rvp.tid); + + /* Only update the transaction ID for new traps (o13-5). */ + if (trap->tid == 0) { + ibp->rvp.tid++; + /* make sure that tid != 0 */ + if (ibp->rvp.tid == 0) + ibp->rvp.tid++; + trap->tid = cpu_to_be64(ibp->rvp.tid); + } + smp->tid = trap->tid; + smp->attr_id = IB_SMP_ATTR_NOTICE; /* o14-1: smp->mkey = 0; */ - memcpy(smp->route.lid.data, data, len); + + memcpy(smp->route.lid.data, &trap->data, trap->len); spin_lock_irqsave(&ibp->rvp.lock, flags); if (!ibp->rvp.sm_ah) { @@ -152,31 +300,72 @@ static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len) ah = hfi1_create_qp0_ah(ibp, ibp->rvp.sm_lid); if (IS_ERR(ah)) { - ret = PTR_ERR(ah); - } else { - send_buf->ah = ah; - ibp->rvp.sm_ah = ibah_to_rvtah(ah); - ret = 0; + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + return; } + send_buf->ah = ah; + ibp->rvp.sm_ah = ibah_to_rvtah(ah); } else { - ret = -EINVAL; + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + return; } } else { send_buf->ah = &ibp->rvp.sm_ah->ibah; - ret = 0; } + + /* + * If the trap was repressed while things were getting set up, don't + * bother sending it. This could happen for a retry. + */ + if (trap->repress) { + list_del(&trap->list); + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + kfree(trap); + ib_free_send_mad(send_buf); + return; + } + + trap->in_use = 0; spin_unlock_irqrestore(&ibp->rvp.lock, flags); - if (!ret) - ret = ib_post_send_mad(send_buf, NULL); - if (!ret) { - /* 4.096 usec. */ - timeout = (4096 * (1UL << ibp->rvp.subnet_timeout)) / 1000; - ibp->rvp.trap_timeout = jiffies + usecs_to_jiffies(timeout); - } else { + if (ib_post_send_mad(send_buf, NULL)) ib_free_send_mad(send_buf); - ibp->rvp.trap_timeout = 0; +} + +void hfi1_handle_trap_timer(unsigned long data) +{ + struct hfi1_ibport *ibp = (struct hfi1_ibport *)data; + struct trap_node *trap = NULL; + unsigned long flags; + int i; + + /* Find the trap with the highest priority */ + spin_lock_irqsave(&ibp->rvp.lock, flags); + for (i = 0; !trap && i < RVT_MAX_TRAP_LISTS; i++) { + trap = list_first_entry_or_null(&ibp->rvp.trap_lists[i].list, + struct trap_node, list); } + spin_unlock_irqrestore(&ibp->rvp.lock, flags); + + if (trap) + send_trap(ibp, trap); +} + +static struct trap_node *create_trap_node(u8 type, __be16 trap_num, u32 lid) +{ + struct trap_node *trap; + + trap = kzalloc(sizeof(*trap), GFP_ATOMIC); + if (!trap) + return NULL; + + INIT_LIST_HEAD(&trap->list); + trap->data.generic_type = type; + trap->data.prod_type_lsb = IB_NOTICE_PROD_CA; + trap->data.trap_num = trap_num; + trap->data.issuer_lid = cpu_to_be32(lid); + + return trap; } /* @@ -185,28 +374,29 @@ static void send_trap(struct hfi1_ibport *ibp, void *data, unsigned len) void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl, u32 qp1, u32 qp2, u16 lid1, u16 lid2) { - struct opa_mad_notice_attr data; + struct trap_node *trap; u32 lid = ppd_from_ibp(ibp)->lid; u32 _lid1 = lid1; u32 _lid2 = lid2; - memset(&data, 0, sizeof(data)); ibp->rvp.n_pkt_drops++; ibp->rvp.pkey_violations++; + trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_P_KEY, + lid); + if (!trap) + return; + /* Send violation trap */ - data.generic_type = IB_NOTICE_TYPE_SECURITY; - data.prod_type_lsb = IB_NOTICE_PROD_CA; - data.trap_num = OPA_TRAP_BAD_P_KEY; - data.issuer_lid = cpu_to_be32(lid); - data.ntc_257_258.lid1 = cpu_to_be32(_lid1); - data.ntc_257_258.lid2 = cpu_to_be32(_lid2); - data.ntc_257_258.key = cpu_to_be32(key); - data.ntc_257_258.sl = sl << 3; - data.ntc_257_258.qp1 = cpu_to_be32(qp1); - data.ntc_257_258.qp2 = cpu_to_be32(qp2); - - send_trap(ibp, &data, sizeof(data)); + trap->data.ntc_257_258.lid1 = cpu_to_be32(_lid1); + trap->data.ntc_257_258.lid2 = cpu_to_be32(_lid2); + trap->data.ntc_257_258.key = cpu_to_be32(key); + trap->data.ntc_257_258.sl = sl << 3; + trap->data.ntc_257_258.qp1 = cpu_to_be32(qp1); + trap->data.ntc_257_258.qp2 = cpu_to_be32(qp2); + + trap->len = sizeof(trap->data); + send_trap(ibp, trap); } /* @@ -215,34 +405,36 @@ void hfi1_bad_pkey(struct hfi1_ibport *ibp, u32 key, u32 sl, static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad, __be64 mkey, __be32 dr_slid, u8 return_path[], u8 hop_cnt) { - struct opa_mad_notice_attr data; + struct trap_node *trap; u32 lid = ppd_from_ibp(ibp)->lid; - memset(&data, 0, sizeof(data)); + trap = create_trap_node(IB_NOTICE_TYPE_SECURITY, OPA_TRAP_BAD_M_KEY, + lid); + if (!trap) + return; + /* Send violation trap */ - data.generic_type = IB_NOTICE_TYPE_SECURITY; - data.prod_type_lsb = IB_NOTICE_PROD_CA; - data.trap_num = OPA_TRAP_BAD_M_KEY; - data.issuer_lid = cpu_to_be32(lid); - data.ntc_256.lid = data.issuer_lid; - data.ntc_256.method = mad->method; - data.ntc_256.attr_id = mad->attr_id; - data.ntc_256.attr_mod = mad->attr_mod; - data.ntc_256.mkey = mkey; + trap->data.ntc_256.lid = trap->data.issuer_lid; + trap->data.ntc_256.method = mad->method; + trap->data.ntc_256.attr_id = mad->attr_id; + trap->data.ntc_256.attr_mod = mad->attr_mod; + trap->data.ntc_256.mkey = mkey; if (mad->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) { - data.ntc_256.dr_slid = dr_slid; - data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE; - if (hop_cnt > ARRAY_SIZE(data.ntc_256.dr_rtn_path)) { - data.ntc_256.dr_trunc_hop |= + trap->data.ntc_256.dr_slid = dr_slid; + trap->data.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE; + if (hop_cnt > ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path)) { + trap->data.ntc_256.dr_trunc_hop |= IB_NOTICE_TRAP_DR_TRUNC; - hop_cnt = ARRAY_SIZE(data.ntc_256.dr_rtn_path); + hop_cnt = ARRAY_SIZE(trap->data.ntc_256.dr_rtn_path); } - data.ntc_256.dr_trunc_hop |= hop_cnt; - memcpy(data.ntc_256.dr_rtn_path, return_path, + trap->data.ntc_256.dr_trunc_hop |= hop_cnt; + memcpy(trap->data.ntc_256.dr_rtn_path, return_path, hop_cnt); } - send_trap(ibp, &data, sizeof(data)); + trap->len = sizeof(trap->data); + + send_trap(ibp, trap); } /* @@ -250,23 +442,24 @@ static void bad_mkey(struct hfi1_ibport *ibp, struct ib_mad_hdr *mad, */ void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num) { - struct opa_mad_notice_attr data; + struct trap_node *trap; struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi); struct hfi1_devdata *dd = dd_from_dev(verbs_dev); struct hfi1_ibport *ibp = &dd->pport[port_num - 1].ibport_data; u32 lid = ppd_from_ibp(ibp)->lid; - memset(&data, 0, sizeof(data)); + trap = create_trap_node(IB_NOTICE_TYPE_INFO, + OPA_TRAP_CHANGE_CAPABILITY, + lid); + if (!trap) + return; - data.generic_type = IB_NOTICE_TYPE_INFO; - data.prod_type_lsb = IB_NOTICE_PROD_CA; - data.trap_num = OPA_TRAP_CHANGE_CAPABILITY; - data.issuer_lid = cpu_to_be32(lid); - data.ntc_144.lid = data.issuer_lid; - data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags); - data.ntc_144.cap_mask3 = cpu_to_be16(ibp->rvp.port_cap3_flags); + trap->data.ntc_144.lid = trap->data.issuer_lid; + trap->data.ntc_144.new_cap_mask = cpu_to_be32(ibp->rvp.port_cap_flags); + trap->data.ntc_144.cap_mask3 = cpu_to_be16(ibp->rvp.port_cap3_flags); - send_trap(ibp, &data, sizeof(data)); + trap->len = sizeof(trap->data); + send_trap(ibp, trap); } /* @@ -274,19 +467,19 @@ void hfi1_cap_mask_chg(struct rvt_dev_info *rdi, u8 port_num) */ void hfi1_sys_guid_chg(struct hfi1_ibport *ibp) { - struct opa_mad_notice_attr data; + struct trap_node *trap; u32 lid = ppd_from_ibp(ibp)->lid; - memset(&data, 0, sizeof(data)); + trap = create_trap_node(IB_NOTICE_TYPE_INFO, OPA_TRAP_CHANGE_SYSGUID, + lid); + if (!trap) + return; - data.generic_type = IB_NOTICE_TYPE_INFO; - data.prod_type_lsb = IB_NOTICE_PROD_CA; - data.trap_num = OPA_TRAP_CHANGE_SYSGUID; - data.issuer_lid = cpu_to_be32(lid); - data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid; - data.ntc_145.lid = data.issuer_lid; + trap->data.ntc_145.new_sys_guid = ib_hfi1_sys_image_guid; + trap->data.ntc_145.lid = trap->data.issuer_lid; - send_trap(ibp, &data, sizeof(data)); + trap->len = sizeof(trap->data); + send_trap(ibp, trap); } /* @@ -294,20 +487,21 @@ void hfi1_sys_guid_chg(struct hfi1_ibport *ibp) */ void hfi1_node_desc_chg(struct hfi1_ibport *ibp) { - struct opa_mad_notice_attr data; + struct trap_node *trap; u32 lid = ppd_from_ibp(ibp)->lid; - memset(&data, 0, sizeof(data)); + trap = create_trap_node(IB_NOTICE_TYPE_INFO, + OPA_TRAP_CHANGE_CAPABILITY, + lid); + if (!trap) + return; - data.generic_type = IB_NOTICE_TYPE_INFO; - data.prod_type_lsb = IB_NOTICE_PROD_CA; - data.trap_num = OPA_TRAP_CHANGE_CAPABILITY; - data.issuer_lid = cpu_to_be32(lid); - data.ntc_144.lid = data.issuer_lid; - data.ntc_144.change_flags = + trap->data.ntc_144.lid = trap->data.issuer_lid; + trap->data.ntc_144.change_flags = cpu_to_be16(OPA_NOTICE_TRAP_NODE_DESC_CHG); - send_trap(ibp, &data, sizeof(data)); + trap->len = sizeof(trap->data); + send_trap(ibp, trap); } static int __subn_get_opa_nodedesc(struct opa_smp *smp, u32 am, @@ -4144,6 +4338,11 @@ static int process_subn_opa(struct ib_device *ibdev, int mad_flags, */ ret = IB_MAD_RESULT_SUCCESS; break; + case IB_MGMT_METHOD_TRAP_REPRESS: + subn_handle_opa_trap_repress(ibp, smp); + /* Always successful */ + ret = IB_MAD_RESULT_SUCCESS; + break; default: smp->status |= IB_SMP_UNSUP_METHOD; ret = reply((struct ib_mad_hdr *)smp); diff --git a/drivers/infiniband/hw/hfi1/mad.h b/drivers/infiniband/hw/hfi1/mad.h index a4e2506bd5ca..4c1245072093 100644 --- a/drivers/infiniband/hw/hfi1/mad.h +++ b/drivers/infiniband/hw/hfi1/mad.h @@ -1,5 +1,5 @@ /* - * Copyright(c) 2015, 2016 Intel Corporation. + * Copyright(c) 2015 - 2017 Intel Corporation. * * This file is provided under a dual BSD/GPLv2 license. When using or * redistributing this file, you may do so under either license. @@ -428,5 +428,6 @@ struct sc2vlnt { COUNTER_MASK(1, 4)) void hfi1_event_pkey_change(struct hfi1_devdata *dd, u8 port); +void hfi1_handle_trap_timer(unsigned long data); #endif /* _HFI1_MAD_H */ diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index 3ef6384eae40..dc51bf247006 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -1535,6 +1535,11 @@ static void init_ibport(struct hfi1_pportdata *ppd) ibp->sc_to_sl[i] = i; } + for (i = 0; i < RVT_MAX_TRAP_LISTS ; i++) + INIT_LIST_HEAD(&ibp->rvp.trap_lists[i].list); + setup_timer(&ibp->rvp.trap_timer, hfi1_handle_trap_timer, + (unsigned long)ibp); + spin_lock_init(&ibp->rvp.lock); /* Set the prefix to the default value (see ch. 4.1.1) */ ibp->rvp.gid_prefix = IB_DEFAULT_GID_PREFIX; diff --git a/include/rdma/rdma_vt.h b/include/rdma/rdma_vt.h index 22fb15ff5e8b..fdfac0fd2f82 100644 --- a/include/rdma/rdma_vt.h +++ b/include/rdma/rdma_vt.h @@ -57,11 +57,21 @@ #include <linux/list.h> #include <linux/hash.h> #include <rdma/ib_verbs.h> +#include <rdma/ib_mad.h> #include <rdma/rdmavt_mr.h> #include <rdma/rdmavt_qp.h> #define RVT_MAX_PKEY_VALUES 16 +#define RVT_MAX_TRAP_LEN 100 /* Limit pending trap list */ +#define RVT_MAX_TRAP_LISTS ((IB_NOTICE_TYPE_INFO & 0x0F) + 1) +#define RVT_TRAP_TIMEOUT 4096 /* 4.096 usec */ + +struct trap_list { + u32 list_len; + struct list_head list; +}; + struct rvt_ibport { struct rvt_qp __rcu *qp[2]; struct ib_mad_agent *send_agent; /* agent for SMI (traps) */ @@ -128,6 +138,13 @@ struct rvt_ibport { u16 *pkey_table; struct rvt_ah *sm_ah; + + /* + * Keep a list of traps that have not been repressed. They will be + * resent based on trap_timer. + */ + struct trap_list trap_lists[RVT_MAX_TRAP_LISTS]; + struct timer_list trap_timer; }; #define RVT_CQN_MAX 16 /* maximum length of cq name */ |