X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=drivers%2Finfiniband%2Fhw%2Fehca%2Fehca_qp.c;h=4dbe2870e0145fac40e3e31e3d8a01298fa10f17;hb=b9012e0a4255c93e1d81f1ccee591de6414b5955;hp=f95403cb523dd3c577ffafc97f56a1da45fcdec3;hpb=b708fba3c2942a175c3cb04a7bb4c89f907b497b;p=safe%2Fjmp%2Flinux-2.6 diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c index f95403c..4dbe287 100644 --- a/drivers/infiniband/hw/ehca/ehca_qp.c +++ b/drivers/infiniband/hw/ehca/ehca_qp.c @@ -43,9 +43,6 @@ * POSSIBILITY OF SUCH DAMAGE. */ - -#include - #include "ehca_classes.h" #include "ehca_tools.h" #include "ehca_qes.h" @@ -399,6 +396,50 @@ static void ehca_determine_small_queue(struct ehca_alloc_queue_parms *queue, queue->is_small = (queue->page_size != 0); } +/* needs to be called with cq->spinlock held */ +void ehca_add_to_err_list(struct ehca_qp *qp, int on_sq) +{ + struct list_head *list, *node; + + /* TODO: support low latency QPs */ + if (qp->ext_type == EQPT_LLQP) + return; + + if (on_sq) { + list = &qp->send_cq->sqp_err_list; + node = &qp->sq_err_node; + } else { + list = &qp->recv_cq->rqp_err_list; + node = &qp->rq_err_node; + } + + if (list_empty(node)) + list_add_tail(node, list); + + return; +} + +static void del_from_err_list(struct ehca_cq *cq, struct list_head *node) +{ + unsigned long flags; + + spin_lock_irqsave(&cq->spinlock, flags); + + if (!list_empty(node)) + list_del_init(node); + + spin_unlock_irqrestore(&cq->spinlock, flags); +} + +static void reset_queue_map(struct ehca_queue_map *qmap) +{ + int i; + + qmap->tail = 0; + for (i = 0; i < qmap->entries; i++) + qmap->map[i].reported = 1; +} + /* * Create an ib_qp struct that is either a QP or an SRQ, depending on * the value of the is_srq parameter. If init_attr and srq_init_attr share @@ -410,7 +451,7 @@ static struct ehca_qp *internal_create_qp( struct ib_srq_init_attr *srq_init_attr, struct ib_udata *udata, int is_srq) { - struct ehca_qp *my_qp; + struct ehca_qp *my_qp, *my_srq = NULL; struct ehca_pd *my_pd = container_of(pd, struct ehca_pd, ib_pd); struct ehca_shca *shca = container_of(pd->device, struct ehca_shca, ib_device); @@ -424,6 +465,19 @@ static struct ehca_qp *internal_create_qp( u32 swqe_size = 0, rwqe_size = 0, ib_qp_num; unsigned long flags; + if (!atomic_add_unless(&shca->num_qps, 1, ehca_max_qp)) { + ehca_err(pd->device, "Unable to create QP, max number of %i " + "QPs reached.", ehca_max_qp); + ehca_err(pd->device, "To increase the maximum number of QPs " + "use the number_of_qps module parameter.\n"); + return ERR_PTR(-ENOSPC); + } + + if (init_attr->create_flags) { + atomic_dec(&shca->num_qps); + return ERR_PTR(-EINVAL); + } + memset(&parms, 0, sizeof(parms)); qp_type = init_attr->qp_type; @@ -431,6 +485,7 @@ static struct ehca_qp *internal_create_qp( init_attr->sq_sig_type != IB_SIGNAL_ALL_WR) { ehca_err(pd->device, "init_attr->sg_sig_type=%x not allowed", init_attr->sq_sig_type); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } @@ -445,17 +500,16 @@ static struct ehca_qp *internal_create_qp( /* handle SRQ base QPs */ if (init_attr->srq) { - struct ehca_qp *my_srq = - container_of(init_attr->srq, struct ehca_qp, ib_srq); + my_srq = container_of(init_attr->srq, struct ehca_qp, ib_srq); has_srq = 1; parms.ext_type = EQPT_SRQBASE; parms.srq_qpn = my_srq->real_qp_num; - parms.srq_token = my_srq->token; } if (is_llqp && has_srq) { ehca_err(pd->device, "LLQPs can't have an SRQ"); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } @@ -467,6 +521,7 @@ static struct ehca_qp *internal_create_qp( ehca_err(pd->device, "no more than three SGEs " "supported for SRQ pd=%p max_sge=%x", pd, init_attr->cap.max_recv_sge); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } } @@ -478,6 +533,7 @@ static struct ehca_qp *internal_create_qp( qp_type != IB_QPT_SMI && qp_type != IB_QPT_GSI) { ehca_err(pd->device, "wrong QP Type=%x", qp_type); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } @@ -491,6 +547,7 @@ static struct ehca_qp *internal_create_qp( "or max_rq_wr=%x for RC LLQP", init_attr->cap.max_send_wr, init_attr->cap.max_recv_wr); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } break; @@ -498,6 +555,7 @@ static struct ehca_qp *internal_create_qp( if (!EHCA_BMASK_GET(HCA_CAP_UD_LL_QP, shca->hca_cap)) { ehca_err(pd->device, "UD LLQP not supported " "by this adapter"); + atomic_dec(&shca->num_qps); return ERR_PTR(-ENOSYS); } if (!(init_attr->cap.max_send_sge <= 5 @@ -509,20 +567,22 @@ static struct ehca_qp *internal_create_qp( "or max_recv_sge=%x for UD LLQP", init_attr->cap.max_send_sge, init_attr->cap.max_recv_sge); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } else if (init_attr->cap.max_send_wr > 255) { ehca_err(pd->device, "Invalid Number of " "max_send_wr=%x for UD QP_TYPE=%x", init_attr->cap.max_send_wr, qp_type); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } break; default: ehca_err(pd->device, "unsupported LL QP Type=%x", qp_type); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); - break; } } else { int max_sge = (qp_type == IB_QPT_UD || qp_type == IB_QPT_SMI @@ -534,6 +594,7 @@ static struct ehca_qp *internal_create_qp( "send_sge=%x recv_sge=%x max_sge=%x", init_attr->cap.max_send_sge, init_attr->cap.max_recv_sge, max_sge); + atomic_dec(&shca->num_qps); return ERR_PTR(-EINVAL); } } @@ -544,13 +605,17 @@ static struct ehca_qp *internal_create_qp( my_qp = kmem_cache_zalloc(qp_cache, GFP_KERNEL); if (!my_qp) { ehca_err(pd->device, "pd=%p not enough memory to alloc qp", pd); + atomic_dec(&shca->num_qps); return ERR_PTR(-ENOMEM); } + atomic_set(&my_qp->nr_events, 0); + init_waitqueue_head(&my_qp->wait_completion); spin_lock_init(&my_qp->spinlock_s); spin_lock_init(&my_qp->spinlock_r); my_qp->qp_type = qp_type; my_qp->ext_type = parms.ext_type; + my_qp->state = IB_QPS_RESET; if (init_attr->recv_cq) my_qp->recv_cq = @@ -583,6 +648,9 @@ static struct ehca_qp *internal_create_qp( goto create_qp_exit1; } + if (has_srq) + parms.srq_token = my_qp->token; + parms.servicetype = ibqptype2servicetype(qp_type); if (parms.servicetype < 0) { ret = -EINVAL; @@ -590,10 +658,8 @@ static struct ehca_qp *internal_create_qp( goto create_qp_exit1; } - if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) - parms.sigtype = HCALL_SIGT_EVERY; - else - parms.sigtype = HCALL_SIGT_BY_WQE; + /* Always signal by WQE so we can hide circ. WQEs */ + parms.sigtype = HCALL_SIGT_BY_WQE; /* UD_AV CIRCUMVENTION */ max_send_sge = init_attr->cap.max_send_sge; @@ -616,6 +682,10 @@ static struct ehca_qp *internal_create_qp( parms.squeue.max_sge = max_send_sge; parms.rqueue.max_sge = max_recv_sge; + /* RC QPs need one more SWQE for unsolicited ack circumvention */ + if (qp_type == IB_QPT_RC) + parms.squeue.max_wr++; + if (EHCA_BMASK_GET(HCA_CAP_MINI_QP, shca->hca_cap)) { if (HAS_SQ(my_qp)) ehca_determine_small_queue( @@ -648,6 +718,8 @@ static struct ehca_qp *internal_create_qp( parms.squeue.act_nr_sges = 1; parms.rqueue.act_nr_sges = 1; } + /* hide the extra WQE */ + parms.squeue.act_nr_wqes--; break; case IB_QPT_UD: case IB_QPT_GSI: @@ -686,6 +758,19 @@ static struct ehca_qp *internal_create_qp( "and pages ret=%i", ret); goto create_qp_exit2; } + + my_qp->sq_map.entries = my_qp->ipz_squeue.queue_length / + my_qp->ipz_squeue.qe_size; + my_qp->sq_map.map = vmalloc(my_qp->sq_map.entries * + sizeof(struct ehca_qmap_entry)); + if (!my_qp->sq_map.map) { + ehca_err(pd->device, "Couldn't allocate squeue " + "map ret=%i", ret); + goto create_qp_exit3; + } + INIT_LIST_HEAD(&my_qp->sq_err_node); + /* to avoid the generation of bogus flush CQEs */ + reset_queue_map(&my_qp->sq_map); } if (HAS_RQ(my_qp)) { @@ -695,8 +780,27 @@ static struct ehca_qp *internal_create_qp( if (ret) { ehca_err(pd->device, "Couldn't initialize rqueue " "and pages ret=%i", ret); - goto create_qp_exit3; + goto create_qp_exit4; } + + my_qp->rq_map.entries = my_qp->ipz_rqueue.queue_length / + my_qp->ipz_rqueue.qe_size; + my_qp->rq_map.map = vmalloc(my_qp->rq_map.entries * + sizeof(struct ehca_qmap_entry)); + if (!my_qp->rq_map.map) { + ehca_err(pd->device, "Couldn't allocate squeue " + "map ret=%i", ret); + goto create_qp_exit5; + } + INIT_LIST_HEAD(&my_qp->rq_err_node); + /* to avoid the generation of bogus flush CQEs */ + reset_queue_map(&my_qp->rq_map); + } else if (init_attr->srq) { + /* this is a base QP, use the queue map of the SRQ */ + my_qp->rq_map = my_srq->rq_map; + INIT_LIST_HEAD(&my_qp->rq_err_node); + + my_qp->ipz_rqueue = my_srq->ipz_rqueue; } if (is_srq) { @@ -727,12 +831,31 @@ static struct ehca_qp *internal_create_qp( init_attr->cap.max_send_wr = parms.squeue.act_nr_wqes; my_qp->init_attr = *init_attr; + if (qp_type == IB_QPT_SMI || qp_type == IB_QPT_GSI) { + shca->sport[init_attr->port_num - 1].ibqp_sqp[qp_type] = + &my_qp->ib_qp; + if (ehca_nr_ports < 0) { + /* alloc array to cache subsequent modify qp parms + * for autodetect mode + */ + my_qp->mod_qp_parm = + kzalloc(EHCA_MOD_QP_PARM_MAX * + sizeof(*my_qp->mod_qp_parm), + GFP_KERNEL); + if (!my_qp->mod_qp_parm) { + ehca_err(pd->device, + "Could not alloc mod_qp_parm"); + goto create_qp_exit5; + } + } + } + /* NOTE: define_apq0() not supported yet */ if (qp_type == IB_QPT_GSI) { h_ret = ehca_define_sqp(shca, my_qp, init_attr); if (h_ret != H_SUCCESS) { ret = ehca2ib_return_code(h_ret); - goto create_qp_exit4; + goto create_qp_exit6; } } @@ -741,7 +864,7 @@ static struct ehca_qp *internal_create_qp( if (ret) { ehca_err(pd->device, "Couldn't assign qp to send_cq ret=%i", ret); - goto create_qp_exit4; + goto create_qp_exit7; } } @@ -767,16 +890,30 @@ static struct ehca_qp *internal_create_qp( if (ib_copy_to_udata(udata, &resp, sizeof resp)) { ehca_err(pd->device, "Copy to udata failed"); ret = -EINVAL; - goto create_qp_exit4; + goto create_qp_exit8; } } return my_qp; -create_qp_exit4: +create_qp_exit8: + ehca_cq_unassign_qp(my_qp->send_cq, my_qp->real_qp_num); + +create_qp_exit7: + kfree(my_qp->mod_qp_parm); + +create_qp_exit6: + if (HAS_RQ(my_qp)) + vfree(my_qp->rq_map.map); + +create_qp_exit5: if (HAS_RQ(my_qp)) ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue); +create_qp_exit4: + if (HAS_SQ(my_qp)) + vfree(my_qp->sq_map.map); + create_qp_exit3: if (HAS_SQ(my_qp)) ipz_queue_dtor(my_pd, &my_qp->ipz_squeue); @@ -791,6 +928,7 @@ create_qp_exit1: create_qp_exit0: kmem_cache_free(qp_cache, my_qp); + atomic_dec(&shca->num_qps); return ERR_PTR(ret); } @@ -836,7 +974,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd, /* copy back return values */ srq_init_attr->attr.max_wr = qp_init_attr.cap.max_recv_wr; - srq_init_attr->attr.max_sge = qp_init_attr.cap.max_recv_sge; + srq_init_attr->attr.max_sge = 3; /* drive SRQ into RTR state */ mqpcb = ehca_alloc_fw_ctrlblock(GFP_KERNEL); @@ -856,7 +994,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd, update_mask, mqpcb, my_qp->galpas.kernel); if (hret != H_SUCCESS) { - ehca_err(pd->device, "Could not modify SRQ to INIT" + ehca_err(pd->device, "Could not modify SRQ to INIT " "ehca_qp=%p qp_num=%x h_ret=%li", my_qp, my_qp->real_qp_num, hret); goto create_srq2; @@ -870,7 +1008,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd, update_mask, mqpcb, my_qp->galpas.kernel); if (hret != H_SUCCESS) { - ehca_err(pd->device, "Could not enable SRQ" + ehca_err(pd->device, "Could not enable SRQ " "ehca_qp=%p qp_num=%x h_ret=%li", my_qp, my_qp->real_qp_num, hret); goto create_srq2; @@ -884,12 +1022,14 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd, update_mask, mqpcb, my_qp->galpas.kernel); if (hret != H_SUCCESS) { - ehca_err(pd->device, "Could not modify SRQ to RTR" + ehca_err(pd->device, "Could not modify SRQ to RTR " "ehca_qp=%p qp_num=%x h_ret=%li", my_qp, my_qp->real_qp_num, hret); goto create_srq2; } + ehca_free_fw_ctrlblock(mqpcb); + return &my_qp->ib_srq; create_srq2: @@ -932,7 +1072,7 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca, qp_num, bad_send_wqe_p); /* convert wqe pointer to vadr */ bad_send_wqe_v = abs_to_virt((u64)bad_send_wqe_p); - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(bad_send_wqe_v, 32, "qp_num=%x bad_wqe", qp_num); squeue = &my_qp->ipz_squeue; if (ipz_queue_abs_to_offset(squeue, (u64)bad_send_wqe_p, &q_ofs)) { @@ -945,7 +1085,7 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca, wqe = (struct ehca_wqe *)ipz_qeit_calc(squeue, q_ofs); *bad_wqe_cnt = 0; while (wqe->optype != 0xff && wqe->wqef != 0xff) { - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(wqe, 32, "qp_num=%x wqe", qp_num); wqe->nr_of_data_seg = 0; /* suppress data access */ wqe->wqef = WQEF_PURGE; /* WQE to be purged */ @@ -964,6 +1104,101 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca, return 0; } +static int calc_left_cqes(u64 wqe_p, struct ipz_queue *ipz_queue, + struct ehca_queue_map *qmap) +{ + void *wqe_v; + u64 q_ofs; + u32 wqe_idx; + + /* convert real to abs address */ + wqe_p = wqe_p & (~(1UL << 63)); + + wqe_v = abs_to_virt(wqe_p); + + if (ipz_queue_abs_to_offset(ipz_queue, wqe_p, &q_ofs)) { + ehca_gen_err("Invalid offset for calculating left cqes " + "wqe_p=%#lx wqe_v=%p\n", wqe_p, wqe_v); + return -EFAULT; + } + + wqe_idx = q_ofs / ipz_queue->qe_size; + if (wqe_idx < qmap->tail) + qmap->left_to_poll = (qmap->entries - qmap->tail) + wqe_idx; + else + qmap->left_to_poll = wqe_idx - qmap->tail; + + return 0; +} + +static int check_for_left_cqes(struct ehca_qp *my_qp, struct ehca_shca *shca) +{ + u64 h_ret; + void *send_wqe_p, *recv_wqe_p; + int ret; + unsigned long flags; + int qp_num = my_qp->ib_qp.qp_num; + + /* this hcall is not supported on base QPs */ + if (my_qp->ext_type != EQPT_SRQBASE) { + /* get send and receive wqe pointer */ + h_ret = hipz_h_disable_and_get_wqe(shca->ipz_hca_handle, + my_qp->ipz_qp_handle, &my_qp->pf, + &send_wqe_p, &recv_wqe_p, 4); + if (h_ret != H_SUCCESS) { + ehca_err(&shca->ib_device, "disable_and_get_wqe() " + "failed ehca_qp=%p qp_num=%x h_ret=%li", + my_qp, qp_num, h_ret); + return ehca2ib_return_code(h_ret); + } + + /* + * acquire lock to ensure that nobody is polling the cq which + * could mean that the qmap->tail pointer is in an + * inconsistent state. + */ + spin_lock_irqsave(&my_qp->send_cq->spinlock, flags); + ret = calc_left_cqes((u64)send_wqe_p, &my_qp->ipz_squeue, + &my_qp->sq_map); + spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags); + if (ret) + return ret; + + + spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags); + ret = calc_left_cqes((u64)recv_wqe_p, &my_qp->ipz_rqueue, + &my_qp->rq_map); + spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags); + if (ret) + return ret; + } else { + spin_lock_irqsave(&my_qp->send_cq->spinlock, flags); + my_qp->sq_map.left_to_poll = 0; + spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags); + + spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags); + my_qp->rq_map.left_to_poll = 0; + spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags); + } + + /* this assures flush cqes being generated only for pending wqes */ + if ((my_qp->sq_map.left_to_poll == 0) && + (my_qp->rq_map.left_to_poll == 0)) { + spin_lock_irqsave(&my_qp->send_cq->spinlock, flags); + ehca_add_to_err_list(my_qp, 1); + spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags); + + if (HAS_RQ(my_qp)) { + spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags); + ehca_add_to_err_list(my_qp, 0); + spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, + flags); + } + } + + return 0; +} + /* * internal_modify_qp with circumvention to handle aqp0 properly * smi_reset2init indicates if this is an internal reset-to-init-call for @@ -988,7 +1223,7 @@ static int internal_modify_qp(struct ib_qp *ibqp, unsigned long flags = 0; /* do query_qp to obtain current attr values */ - mqpcb = ehca_alloc_fw_ctrlblock(GFP_KERNEL); + mqpcb = ehca_alloc_fw_ctrlblock(GFP_ATOMIC); if (!mqpcb) { ehca_err(ibqp->device, "Could not get zeroed page for mqpcb " "ehca_qp=%p qp_num=%x ", my_qp, ibqp->qp_num); @@ -1165,10 +1400,19 @@ static int internal_modify_qp(struct ib_qp *ibqp, } if (attr_mask & IB_QP_PKEY_INDEX) { + if (attr->pkey_index >= 16) { + ret = -EINVAL; + ehca_err(ibqp->device, "Invalid pkey_index=%x. " + "ehca_qp=%p qp_num=%x max_pkey_index=f", + attr->pkey_index, my_qp, ibqp->qp_num); + goto modify_qp_exit2; + } mqpcb->prim_p_key_idx = attr->pkey_index; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_PRIM_P_KEY_IDX, 1); } if (attr_mask & IB_QP_PORT) { + struct ehca_sport *sport; + struct ehca_qp *aqp1; if (attr->port_num < 1 || attr->port_num > shca->num_ports) { ret = -EINVAL; ehca_err(ibqp->device, "Invalid port=%x. " @@ -1177,6 +1421,29 @@ static int internal_modify_qp(struct ib_qp *ibqp, shca->num_ports); goto modify_qp_exit2; } + sport = &shca->sport[attr->port_num - 1]; + if (!sport->ibqp_sqp[IB_QPT_GSI]) { + /* should not occur */ + ret = -EFAULT; + ehca_err(ibqp->device, "AQP1 was not created for " + "port=%x", attr->port_num); + goto modify_qp_exit2; + } + aqp1 = container_of(sport->ibqp_sqp[IB_QPT_GSI], + struct ehca_qp, ib_qp); + if (ibqp->qp_type != IB_QPT_GSI && + ibqp->qp_type != IB_QPT_SMI && + aqp1->mod_qp_parm) { + /* + * firmware will reject this modify_qp() because + * port is not activated/initialized fully + */ + ret = -EFAULT; + ehca_warn(ibqp->device, "Couldn't modify qp port=%x: " + "either port is being activated (try again) " + "or cabling issue", attr->port_num); + goto modify_qp_exit2; + } mqpcb->prim_phys_port = attr->port_num; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_PRIM_PHYS_PORT, 1); } @@ -1185,10 +1452,6 @@ static int internal_modify_qp(struct ib_qp *ibqp, update_mask |= EHCA_BMASK_SET(MQPCB_MASK_QKEY, 1); } if (attr_mask & IB_QP_AV) { - int ah_mult = ib_rate_to_mult(attr->ah_attr.static_rate); - int ehca_mult = ib_rate_to_mult(shca->sport[my_qp-> - init_attr.port_num].rate); - mqpcb->dlid = attr->ah_attr.dlid; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_DLID, 1); mqpcb->source_path_bits = attr->ah_attr.src_path_bits; @@ -1196,11 +1459,12 @@ static int internal_modify_qp(struct ib_qp *ibqp, mqpcb->service_level = attr->ah_attr.sl; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_SERVICE_LEVEL, 1); - if (ah_mult < ehca_mult) - mqpcb->max_static_rate = (ah_mult > 0) ? - ((ehca_mult - 1) / ah_mult) : 0; - else - mqpcb->max_static_rate = 0; + if (ehca_calc_ipd(shca, mqpcb->prim_phys_port, + attr->ah_attr.static_rate, + &mqpcb->max_static_rate)) { + ret = -EINVAL; + goto modify_qp_exit2; + } update_mask |= EHCA_BMASK_SET(MQPCB_MASK_MAX_STATIC_RATE, 1); /* @@ -1236,6 +1500,8 @@ static int internal_modify_qp(struct ib_qp *ibqp, } if (attr_mask & IB_QP_PATH_MTU) { + /* store ld(MTU) */ + my_qp->mtu_shift = attr->path_mtu + 7; mqpcb->path_mtu = attr->path_mtu; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_PATH_MTU, 1); } @@ -1269,54 +1535,80 @@ static int internal_modify_qp(struct ib_qp *ibqp, (MQPCB_MASK_RDMA_ATOMIC_OUTST_DEST_QP, 1); } if (attr_mask & IB_QP_ALT_PATH) { - int ah_mult = ib_rate_to_mult(attr->alt_ah_attr.static_rate); - int ehca_mult = ib_rate_to_mult( - shca->sport[my_qp->init_attr.port_num].rate); + if (attr->alt_port_num < 1 + || attr->alt_port_num > shca->num_ports) { + ret = -EINVAL; + ehca_err(ibqp->device, "Invalid alt_port=%x. " + "ehca_qp=%p qp_num=%x num_ports=%x", + attr->alt_port_num, my_qp, ibqp->qp_num, + shca->num_ports); + goto modify_qp_exit2; + } + mqpcb->alt_phys_port = attr->alt_port_num; + if (attr->alt_pkey_index >= 16) { + ret = -EINVAL; + ehca_err(ibqp->device, "Invalid alt_pkey_index=%x. " + "ehca_qp=%p qp_num=%x max_pkey_index=f", + attr->pkey_index, my_qp, ibqp->qp_num); + goto modify_qp_exit2; + } + mqpcb->alt_p_key_idx = attr->alt_pkey_index; + + mqpcb->timeout_al = attr->alt_timeout; mqpcb->dlid_al = attr->alt_ah_attr.dlid; - update_mask |= EHCA_BMASK_SET(MQPCB_MASK_DLID_AL, 1); mqpcb->source_path_bits_al = attr->alt_ah_attr.src_path_bits; - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_SOURCE_PATH_BITS_AL, 1); mqpcb->service_level_al = attr->alt_ah_attr.sl; - update_mask |= EHCA_BMASK_SET(MQPCB_MASK_SERVICE_LEVEL_AL, 1); - if (ah_mult < ehca_mult) - mqpcb->max_static_rate = (ah_mult > 0) ? - ((ehca_mult - 1) / ah_mult) : 0; - else - mqpcb->max_static_rate_al = 0; + if (ehca_calc_ipd(shca, mqpcb->alt_phys_port, + attr->alt_ah_attr.static_rate, + &mqpcb->max_static_rate_al)) { + ret = -EINVAL; + goto modify_qp_exit2; + } + + /* OpenIB doesn't support alternate retry counts - copy them */ + mqpcb->retry_count_al = mqpcb->retry_count; + mqpcb->rnr_retry_count_al = mqpcb->rnr_retry_count; - update_mask |= EHCA_BMASK_SET(MQPCB_MASK_MAX_STATIC_RATE_AL, 1); + update_mask |= EHCA_BMASK_SET(MQPCB_MASK_ALT_PHYS_PORT, 1) + | EHCA_BMASK_SET(MQPCB_MASK_ALT_P_KEY_IDX, 1) + | EHCA_BMASK_SET(MQPCB_MASK_TIMEOUT_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_DLID_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_SOURCE_PATH_BITS_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_SERVICE_LEVEL_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_MAX_STATIC_RATE_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_RETRY_COUNT_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_RNR_RETRY_COUNT_AL, 1); + + /* + * Always supply the GRH flag, even if it's zero, to give the + * hypervisor a clear "yes" or "no" instead of a "perhaps" + */ + update_mask |= EHCA_BMASK_SET(MQPCB_MASK_SEND_GRH_FLAG_AL, 1); /* * only if GRH is TRUE we might consider SOURCE_GID_IDX * and DEST_GID otherwise phype will return H_ATTR_PARM!!! */ if (attr->alt_ah_attr.ah_flags == IB_AH_GRH) { - mqpcb->send_grh_flag_al = 1 << 31; - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_SEND_GRH_FLAG_AL, 1); - mqpcb->source_gid_idx_al = - attr->alt_ah_attr.grh.sgid_index; - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_SOURCE_GID_IDX_AL, 1); + mqpcb->send_grh_flag_al = 1; for (cnt = 0; cnt < 16; cnt++) mqpcb->dest_gid_al.byte[cnt] = attr->alt_ah_attr.grh.dgid.raw[cnt]; - - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_DEST_GID_AL, 1); + mqpcb->source_gid_idx_al = + attr->alt_ah_attr.grh.sgid_index; mqpcb->flow_label_al = attr->alt_ah_attr.grh.flow_label; - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_FLOW_LABEL_AL, 1); mqpcb->hop_limit_al = attr->alt_ah_attr.grh.hop_limit; - update_mask |= - EHCA_BMASK_SET(MQPCB_MASK_HOP_LIMIT_AL, 1); mqpcb->traffic_class_al = attr->alt_ah_attr.grh.traffic_class; + update_mask |= + EHCA_BMASK_SET(MQPCB_MASK_SOURCE_GID_IDX_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_DEST_GID_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_FLOW_LABEL_AL, 1) + | EHCA_BMASK_SET(MQPCB_MASK_HOP_LIMIT_AL, 1) | EHCA_BMASK_SET(MQPCB_MASK_TRAFFIC_CLASS_AL, 1); } } @@ -1338,7 +1630,16 @@ static int internal_modify_qp(struct ib_qp *ibqp, } if (attr_mask & IB_QP_PATH_MIG_STATE) { - mqpcb->path_migration_state = attr->path_mig_state; + if (attr->path_mig_state != IB_MIG_REARM + && attr->path_mig_state != IB_MIG_MIGRATED) { + ret = -EINVAL; + ehca_err(ibqp->device, "Invalid mig_state=%x", + attr->path_mig_state); + goto modify_qp_exit2; + } + mqpcb->path_migration_state = attr->path_mig_state + 1; + if (attr->path_mig_state == IB_MIG_REARM) + my_qp->mig_armed = 1; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_PATH_MIGRATION_STATE, 1); } @@ -1353,7 +1654,7 @@ static int internal_modify_qp(struct ib_qp *ibqp, /* no support for max_send/recv_sge yet */ } - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(mqpcb, 4*70, "qp_num=%x", ibqp->qp_num); h_ret = hipz_h_modify_qp(shca->ipz_hca_handle, @@ -1402,10 +1703,27 @@ static int internal_modify_qp(struct ib_qp *ibqp, goto modify_qp_exit2; } } + if ((qp_new_state == IB_QPS_ERR) && (qp_cur_state != IB_QPS_ERR)) { + ret = check_for_left_cqes(my_qp, shca); + if (ret) + goto modify_qp_exit2; + } if (statetrans == IB_QPST_ANY2RESET) { ipz_qeit_reset(&my_qp->ipz_rqueue); ipz_qeit_reset(&my_qp->ipz_squeue); + + if (qp_cur_state == IB_QPS_ERR) { + del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node); + + if (HAS_RQ(my_qp)) + del_from_err_list(my_qp->recv_cq, + &my_qp->rq_err_node); + } + reset_queue_map(&my_qp->sq_map); + + if (HAS_RQ(my_qp)) + reset_queue_map(&my_qp->rq_map); } if (attr_mask & IB_QP_QKEY) @@ -1426,19 +1744,110 @@ modify_qp_exit1: int ehca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask, struct ib_udata *udata) { + int ret = 0; + + struct ehca_shca *shca = container_of(ibqp->device, struct ehca_shca, + ib_device); struct ehca_qp *my_qp = container_of(ibqp, struct ehca_qp, ib_qp); - struct ehca_pd *my_pd = container_of(my_qp->ib_qp.pd, struct ehca_pd, - ib_pd); - u32 cur_pid = current->tgid; - if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context && - my_pd->ownpid != cur_pid) { - ehca_err(ibqp->pd->device, "Invalid caller pid=%x ownpid=%x", - cur_pid, my_pd->ownpid); - return -EINVAL; + /* The if-block below caches qp_attr to be modified for GSI and SMI + * qps during the initialization by ib_mad. When the respective port + * is activated, ie we got an event PORT_ACTIVE, we'll replay the + * cached modify calls sequence, see ehca_recover_sqs() below. + * Why that is required: + * 1) If one port is connected, older code requires that port one + * to be connected and module option nr_ports=1 to be given by + * user, which is very inconvenient for end user. + * 2) Firmware accepts modify_qp() only if respective port has become + * active. Older code had a wait loop of 30sec create_qp()/ + * define_aqp1(), which is not appropriate in practice. This + * code now removes that wait loop, see define_aqp1(), and always + * reports all ports to ib_mad resp. users. Only activated ports + * will then usable for the users. + */ + if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI) { + int port = my_qp->init_attr.port_num; + struct ehca_sport *sport = &shca->sport[port - 1]; + unsigned long flags; + spin_lock_irqsave(&sport->mod_sqp_lock, flags); + /* cache qp_attr only during init */ + if (my_qp->mod_qp_parm) { + struct ehca_mod_qp_parm *p; + if (my_qp->mod_qp_parm_idx >= EHCA_MOD_QP_PARM_MAX) { + ehca_err(&shca->ib_device, + "mod_qp_parm overflow state=%x port=%x" + " type=%x", attr->qp_state, + my_qp->init_attr.port_num, + ibqp->qp_type); + spin_unlock_irqrestore(&sport->mod_sqp_lock, + flags); + return -EINVAL; + } + p = &my_qp->mod_qp_parm[my_qp->mod_qp_parm_idx]; + p->mask = attr_mask; + p->attr = *attr; + my_qp->mod_qp_parm_idx++; + ehca_dbg(&shca->ib_device, + "Saved qp_attr for state=%x port=%x type=%x", + attr->qp_state, my_qp->init_attr.port_num, + ibqp->qp_type); + spin_unlock_irqrestore(&sport->mod_sqp_lock, flags); + goto out; + } + spin_unlock_irqrestore(&sport->mod_sqp_lock, flags); + } + + ret = internal_modify_qp(ibqp, attr, attr_mask, 0); + +out: + if ((ret == 0) && (attr_mask & IB_QP_STATE)) + my_qp->state = attr->qp_state; + + return ret; +} + +void ehca_recover_sqp(struct ib_qp *sqp) +{ + struct ehca_qp *my_sqp = container_of(sqp, struct ehca_qp, ib_qp); + int port = my_sqp->init_attr.port_num; + struct ib_qp_attr attr; + struct ehca_mod_qp_parm *qp_parm; + int i, qp_parm_idx, ret; + unsigned long flags, wr_cnt; + + if (!my_sqp->mod_qp_parm) + return; + ehca_dbg(sqp->device, "SQP port=%x qp_num=%x", port, sqp->qp_num); + + qp_parm = my_sqp->mod_qp_parm; + qp_parm_idx = my_sqp->mod_qp_parm_idx; + for (i = 0; i < qp_parm_idx; i++) { + attr = qp_parm[i].attr; + ret = internal_modify_qp(sqp, &attr, qp_parm[i].mask, 0); + if (ret) { + ehca_err(sqp->device, "Could not modify SQP port=%x " + "qp_num=%x ret=%x", port, sqp->qp_num, ret); + goto free_qp_parm; + } + ehca_dbg(sqp->device, "SQP port=%x qp_num=%x in state=%x", + port, sqp->qp_num, attr.qp_state); + } + + /* re-trigger posted recv wrs */ + wr_cnt = my_sqp->ipz_rqueue.current_q_offset / + my_sqp->ipz_rqueue.qe_size; + if (wr_cnt) { + spin_lock_irqsave(&my_sqp->spinlock_r, flags); + hipz_update_rqa(my_sqp, wr_cnt); + spin_unlock_irqrestore(&my_sqp->spinlock_r, flags); + ehca_dbg(sqp->device, "doorbell port=%x qp_num=%x wr_cnt=%lx", + port, sqp->qp_num, wr_cnt); } - return internal_modify_qp(ibqp, attr, attr_mask, 0); +free_qp_parm: + kfree(qp_parm); + /* this prevents subsequent calls to modify_qp() to cache qp_attr */ + my_sqp->mod_qp_parm = NULL; } int ehca_query_qp(struct ib_qp *qp, @@ -1446,23 +1855,13 @@ int ehca_query_qp(struct ib_qp *qp, int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr) { struct ehca_qp *my_qp = container_of(qp, struct ehca_qp, ib_qp); - struct ehca_pd *my_pd = container_of(my_qp->ib_qp.pd, struct ehca_pd, - ib_pd); struct ehca_shca *shca = container_of(qp->device, struct ehca_shca, ib_device); struct ipz_adapter_handle adapter_handle = shca->ipz_hca_handle; struct hcp_modify_qp_control_block *qpcb; - u32 cur_pid = current->tgid; int cnt, ret = 0; u64 h_ret; - if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context && - my_pd->ownpid != cur_pid) { - ehca_err(qp->device, "Invalid caller pid=%x ownpid=%x", - cur_pid, my_pd->ownpid); - return -EINVAL; - } - if (qp_attr_mask & QP_ATTR_QUERY_NOT_SUPPORTED) { ehca_err(qp->device, "Invalid attribute mask " "ehca_qp=%p qp_num=%x qp_attr_mask=%x ", @@ -1506,7 +1905,7 @@ int ehca_query_qp(struct ib_qp *qp, qp_attr->qkey = qpcb->qkey; qp_attr->path_mtu = qpcb->path_mtu; - qp_attr->path_mig_state = qpcb->path_migration_state; + qp_attr->path_mig_state = qpcb->path_migration_state - 1; qp_attr->rq_psn = qpcb->receive_psn; qp_attr->sq_psn = qpcb->send_psn; qp_attr->min_rnr_timer = qpcb->min_rnr_nak_timer_field; @@ -1593,7 +1992,7 @@ int ehca_query_qp(struct ib_qp *qp, if (qp_init_attr) *qp_init_attr = my_qp->init_attr; - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(qpcb, 4*70, "qp_num=%x", qp->qp_num); query_qp_exit1: @@ -1607,8 +2006,6 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, { struct ehca_qp *my_qp = container_of(ibsrq, struct ehca_qp, ib_srq); - struct ehca_pd *my_pd = - container_of(ibsrq->pd, struct ehca_pd, ib_pd); struct ehca_shca *shca = container_of(ibsrq->pd->device, struct ehca_shca, ib_device); struct hcp_modify_qp_control_block *mqpcb; @@ -1616,14 +2013,6 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, u64 h_ret; int ret = 0; - u32 cur_pid = current->tgid; - if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context && - my_pd->ownpid != cur_pid) { - ehca_err(ibsrq->pd->device, "Invalid caller pid=%x ownpid=%x", - cur_pid, my_pd->ownpid); - return -EINVAL; - } - mqpcb = ehca_alloc_fw_ctrlblock(GFP_KERNEL); if (!mqpcb) { ehca_err(ibsrq->device, "Could not get zeroed page for mqpcb " @@ -1651,7 +2040,7 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr, goto modify_srq_exit0; } - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(mqpcb, 4*70, "qp_num=%x", my_qp->real_qp_num); h_ret = hipz_h_modify_qp(shca->ipz_hca_handle, my_qp->ipz_qp_handle, @@ -1674,22 +2063,13 @@ modify_srq_exit0: int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr) { struct ehca_qp *my_qp = container_of(srq, struct ehca_qp, ib_srq); - struct ehca_pd *my_pd = container_of(srq->pd, struct ehca_pd, ib_pd); struct ehca_shca *shca = container_of(srq->device, struct ehca_shca, ib_device); struct ipz_adapter_handle adapter_handle = shca->ipz_hca_handle; struct hcp_modify_qp_control_block *qpcb; - u32 cur_pid = current->tgid; int ret = 0; u64 h_ret; - if (my_pd->ib_pd.uobject && my_pd->ib_pd.uobject->context && - my_pd->ownpid != cur_pid) { - ehca_err(srq->device, "Invalid caller pid=%x ownpid=%x", - cur_pid, my_pd->ownpid); - return -EINVAL; - } - qpcb = ehca_alloc_fw_ctrlblock(GFP_KERNEL); if (!qpcb) { ehca_err(srq->device, "Out of memory for qpcb " @@ -1709,10 +2089,11 @@ int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr) } srq_attr->max_wr = qpcb->max_nr_outst_recv_wr - 1; + srq_attr->max_sge = 3; srq_attr->srq_limit = EHCA_BMASK_GET( MQPCB_CURR_SRQ_LIMIT, qpcb->curr_srq_limit); - if (ehca_debug_level) + if (ehca_debug_level >= 2) ehca_dmp(qpcb, 4*70, "qp_num=%x", my_qp->real_qp_num); query_srq_exit1: @@ -1727,7 +2108,7 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp, struct ehca_shca *shca = container_of(dev, struct ehca_shca, ib_device); struct ehca_pd *my_pd = container_of(my_qp->ib_qp.pd, struct ehca_pd, ib_pd); - u32 cur_pid = current->tgid; + struct ehca_sport *sport = &shca->sport[my_qp->init_attr.port_num - 1]; u32 qp_num = my_qp->real_qp_num; int ret; u64 h_ret; @@ -1742,11 +2123,6 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp, "user space qp_num=%x", qp_num); return -EINVAL; } - if (my_pd->ownpid != cur_pid) { - ehca_err(dev, "Invalid caller pid=%x ownpid=%x", - cur_pid, my_pd->ownpid); - return -EINVAL; - } } if (my_qp->send_cq) { @@ -1763,6 +2139,19 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp, idr_remove(&ehca_qp_idr, my_qp->token); write_unlock_irqrestore(&ehca_qp_idr_lock, flags); + /* + * SRQs will never get into an error list and do not have a recv_cq, + * so we need to skip them here. + */ + if (HAS_RQ(my_qp) && !IS_SRQ(my_qp)) + del_from_err_list(my_qp->recv_cq, &my_qp->rq_err_node); + + if (HAS_SQ(my_qp)) + del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node); + + /* now wait until all pending events have completed */ + wait_event(my_qp->wait_completion, !atomic_read(&my_qp->nr_events)); + h_ret = hipz_h_destroy_qp(shca->ipz_hca_handle, my_qp); if (h_ret != H_SUCCESS) { ehca_err(dev, "hipz_h_destroy_qp() failed h_ret=%li " @@ -1773,11 +2162,19 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp, port_num = my_qp->init_attr.port_num; qp_type = my_qp->init_attr.qp_type; + if (qp_type == IB_QPT_SMI || qp_type == IB_QPT_GSI) { + spin_lock_irqsave(&sport->mod_sqp_lock, flags); + kfree(my_qp->mod_qp_parm); + my_qp->mod_qp_parm = NULL; + shca->sport[port_num - 1].ibqp_sqp[qp_type] = NULL; + spin_unlock_irqrestore(&sport->mod_sqp_lock, flags); + } + /* no support for IB_QPT_SMI yet */ if (qp_type == IB_QPT_GSI) { struct ib_event event; ehca_info(dev, "device %s: port %x is inactive.", - shca->ib_device.name, port_num); + shca->ib_device.name, port_num); event.device = &shca->ib_device; event.event = IB_EVENT_PORT_ERR; event.element.port_num = port_num; @@ -1785,11 +2182,18 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp, ib_dispatch_event(&event); } - if (HAS_RQ(my_qp)) + if (HAS_RQ(my_qp)) { ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue); - if (HAS_SQ(my_qp)) + + vfree(my_qp->rq_map.map); + } + if (HAS_SQ(my_qp)) { ipz_queue_dtor(my_pd, &my_qp->ipz_squeue); + + vfree(my_qp->sq_map.map); + } kmem_cache_free(qp_cache, my_qp); + atomic_dec(&shca->num_qps); return 0; }