drivers/infiniband/hw/cxgb4/ev.c

   1 /*
   2  * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  */
  32 #include <linux/slab.h>
  33 #include <linux/mman.h>
  34 #include <net/sock.h>
  35
  36 #include "iw_cxgb4.h"
  37
  38 static void print_tpte(struct c4iw_dev *dev, u32 stag)
  39 {
  40         int ret;
  41         struct fw_ri_tpte tpte;
  42
  43         ret = cxgb4_read_tpte(dev->rdev.lldi.ports[0], stag,
  44                               (__be32 *)&tpte);
  45         if (ret) {
  46                 dev_err(&dev->rdev.lldi.pdev->dev,
  47                         "%s cxgb4_read_tpte err %d\n", __func__, ret);
  48                 return;
  49         }
  50         pr_debug("stag idx 0x%x valid %d key 0x%x state %d pdid %d perm 0x%x ps %d len 0x%llx va 0x%llx\n",
  51                  stag & 0xffffff00,
  52                  FW_RI_TPTE_VALID_G(ntohl(tpte.valid_to_pdid)),
  53                  FW_RI_TPTE_STAGKEY_G(ntohl(tpte.valid_to_pdid)),
  54                  FW_RI_TPTE_STAGSTATE_G(ntohl(tpte.valid_to_pdid)),
  55                  FW_RI_TPTE_PDID_G(ntohl(tpte.valid_to_pdid)),
  56                  FW_RI_TPTE_PERM_G(ntohl(tpte.locread_to_qpid)),
  57                  FW_RI_TPTE_PS_G(ntohl(tpte.locread_to_qpid)),
  58                  ((u64)ntohl(tpte.len_hi) << 32) | ntohl(tpte.len_lo),
  59                  ((u64)ntohl(tpte.va_hi) << 32) | ntohl(tpte.va_lo_fbo));
  60 }
  61
  62 static void dump_err_cqe(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
  63 {
  64         __be64 *p = (void *)err_cqe;
  65
  66         dev_err(&dev->rdev.lldi.pdev->dev,
  67                 "AE qpid %d opcode %d status 0x%x "
  68                 "type %d len 0x%x wrid.hi 0x%x wrid.lo 0x%x\n",
  69                 CQE_QPID(err_cqe), CQE_OPCODE(err_cqe),
  70                 CQE_STATUS(err_cqe), CQE_TYPE(err_cqe), ntohl(err_cqe->len),
  71                 CQE_WRID_HI(err_cqe), CQE_WRID_LOW(err_cqe));
  72
  73         pr_debug("%016llx %016llx %016llx %016llx\n",
  74                  be64_to_cpu(p[0]), be64_to_cpu(p[1]), be64_to_cpu(p[2]),
  75                  be64_to_cpu(p[3]));
  76
  77         /*
  78          * Ingress WRITE and READ_RESP errors provide
  79          * the offending stag, so parse and log it.
  80          */
  81         if (RQ_TYPE(err_cqe) && (CQE_OPCODE(err_cqe) == FW_RI_RDMA_WRITE ||
  82                                  CQE_OPCODE(err_cqe) == FW_RI_READ_RESP))
  83                 print_tpte(dev, CQE_WRID_STAG(err_cqe));
  84 }
  85
  86 static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
  87                           struct c4iw_qp *qhp,
  88                           struct t4_cqe *err_cqe,
  89                           enum ib_event_type ib_event)
  90 {
  91         struct ib_event event;
  92         struct c4iw_qp_attributes attrs;
  93         unsigned long flag;
  94
  95         dump_err_cqe(dev, err_cqe);
  96
  97         if (qhp->attr.state == C4IW_QP_STATE_RTS) {
  98                 attrs.next_state = C4IW_QP_STATE_TERMINATE;
  99                 c4iw_modify_qp(qhp->rhp, qhp, C4IW_QP_ATTR_NEXT_STATE,
 100                                &attrs, 0);
 101         }
 102
 103         event.event = ib_event;
 104         event.device = chp->ibcq.device;
 105         if (ib_event == IB_EVENT_CQ_ERR)
 106                 event.element.cq = &chp->ibcq;
 107         else
 108                 event.element.qp = &qhp->ibqp;
 109         if (qhp->ibqp.event_handler)
 110                 (*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
 111
 112         spin_lock_irqsave(&chp->comp_handler_lock, flag);
 113         (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
 114         spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 115 }
 116
 117 void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
 118 {
 119         struct c4iw_cq *chp;
 120         struct c4iw_qp *qhp;
 121         u32 cqid;
 122
 123         spin_lock_irq(&dev->lock);
 124         qhp = get_qhp(dev, CQE_QPID(err_cqe));
 125         if (!qhp) {
 126                 pr_err("BAD AE qpid 0x%x opcode %d status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
 127                        CQE_QPID(err_cqe),
 128                        CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
 129                        CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
 130                        CQE_WRID_LOW(err_cqe));
 131                 spin_unlock_irq(&dev->lock);
 132                 goto out;
 133         }
 134
 135         if (SQ_TYPE(err_cqe))
 136                 cqid = qhp->attr.scq;
 137         else
 138                 cqid = qhp->attr.rcq;
 139         chp = get_chp(dev, cqid);
 140         if (!chp) {
 141                 pr_err("BAD AE cqid 0x%x qpid 0x%x opcode %d status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
 142                        cqid, CQE_QPID(err_cqe),
 143                        CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
 144                        CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
 145                        CQE_WRID_LOW(err_cqe));
 146                 spin_unlock_irq(&dev->lock);
 147                 goto out;
 148         }
 149
 150         c4iw_qp_add_ref(&qhp->ibqp);
 151         atomic_inc(&chp->refcnt);
 152         spin_unlock_irq(&dev->lock);
 153
 154         /* Bad incoming write */
 155         if (RQ_TYPE(err_cqe) &&
 156             (CQE_OPCODE(err_cqe) == FW_RI_RDMA_WRITE)) {
 157                 post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_REQ_ERR);
 158                 goto done;
 159         }
 160
 161         switch (CQE_STATUS(err_cqe)) {
 162
 163         /* Completion Events */
 164         case T4_ERR_SUCCESS:
 165                 pr_err("AE with status 0!\n");
 166                 break;
 167
 168         case T4_ERR_STAG:
 169         case T4_ERR_PDID:
 170         case T4_ERR_QPID:
 171         case T4_ERR_ACCESS:
 172         case T4_ERR_WRAP:
 173         case T4_ERR_BOUND:
 174         case T4_ERR_INVALIDATE_SHARED_MR:
 175         case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
 176                 post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_ACCESS_ERR);
 177                 break;
 178
 179         /* Device Fatal Errors */
 180         case T4_ERR_ECC:
 181         case T4_ERR_ECC_PSTAG:
 182         case T4_ERR_INTERNAL_ERR:
 183                 post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_DEVICE_FATAL);
 184                 break;
 185
 186         /* QP Fatal Errors */
 187         case T4_ERR_OUT_OF_RQE:
 188         case T4_ERR_PBL_ADDR_BOUND:
 189         case T4_ERR_CRC:
 190         case T4_ERR_MARKER:
 191         case T4_ERR_PDU_LEN_ERR:
 192         case T4_ERR_DDP_VERSION:
 193         case T4_ERR_RDMA_VERSION:
 194         case T4_ERR_OPCODE:
 195         case T4_ERR_DDP_QUEUE_NUM:
 196         case T4_ERR_MSN:
 197         case T4_ERR_TBIT:
 198         case T4_ERR_MO:
 199         case T4_ERR_MSN_GAP:
 200         case T4_ERR_MSN_RANGE:
 201         case T4_ERR_RQE_ADDR_BOUND:
 202         case T4_ERR_IRD_OVERFLOW:
 203                 post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
 204                 break;
 205
 206         default:
 207                 pr_err("Unknown T4 status 0x%x QPID 0x%x\n",
 208                        CQE_STATUS(err_cqe), qhp->wq.sq.qid);
 209                 post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
 210                 break;
 211         }
 212 done:
 213         if (atomic_dec_and_test(&chp->refcnt))
 214                 wake_up(&chp->wait);
 215         c4iw_qp_rem_ref(&qhp->ibqp);
 216 out:
 217         return;
 218 }
 219
 220 int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid)
 221 {
 222         struct c4iw_cq *chp;
 223         unsigned long flag;
 224
 225         spin_lock_irqsave(&dev->lock, flag);
 226         chp = get_chp(dev, qid);
 227         if (chp) {
 228                 atomic_inc(&chp->refcnt);
 229                 spin_unlock_irqrestore(&dev->lock, flag);
 230                 t4_clear_cq_armed(&chp->cq);
 231                 spin_lock_irqsave(&chp->comp_handler_lock, flag);
 232                 (*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
 233                 spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 234                 if (atomic_dec_and_test(&chp->refcnt))
 235                         wake_up(&chp->wait);
 236         } else {
 237                 pr_debug("%s unknown cqid 0x%x\n", __func__, qid);
 238                 spin_unlock_irqrestore(&dev->lock, flag);
 239         }
 240         return 0;
 241 }