linux/drivers/infiniband/hw/hfi1/driver.c
<<
>>
Prefs
   1/*
   2 * Copyright(c) 2015, 2016 Intel Corporation.
   3 *
   4 * This file is provided under a dual BSD/GPLv2 license.  When using or
   5 * redistributing this file, you may do so under either license.
   6 *
   7 * GPL LICENSE SUMMARY
   8 *
   9 * This program is free software; you can redistribute it and/or modify
  10 * it under the terms of version 2 of the GNU General Public License as
  11 * published by the Free Software Foundation.
  12 *
  13 * This program is distributed in the hope that it will be useful, but
  14 * WITHOUT ANY WARRANTY; without even the implied warranty of
  15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  16 * General Public License for more details.
  17 *
  18 * BSD LICENSE
  19 *
  20 * Redistribution and use in source and binary forms, with or without
  21 * modification, are permitted provided that the following conditions
  22 * are met:
  23 *
  24 *  - Redistributions of source code must retain the above copyright
  25 *    notice, this list of conditions and the following disclaimer.
  26 *  - Redistributions in binary form must reproduce the above copyright
  27 *    notice, this list of conditions and the following disclaimer in
  28 *    the documentation and/or other materials provided with the
  29 *    distribution.
  30 *  - Neither the name of Intel Corporation nor the names of its
  31 *    contributors may be used to endorse or promote products derived
  32 *    from this software without specific prior written permission.
  33 *
  34 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  35 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  36 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  37 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  38 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  39 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  40 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  41 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  42 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  43 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  44 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  45 *
  46 */
  47
  48#include <linux/spinlock.h>
  49#include <linux/pci.h>
  50#include <linux/io.h>
  51#include <linux/delay.h>
  52#include <linux/netdevice.h>
  53#include <linux/vmalloc.h>
  54#include <linux/module.h>
  55#include <linux/prefetch.h>
  56#include <rdma/ib_verbs.h>
  57
  58#include "hfi.h"
  59#include "trace.h"
  60#include "qp.h"
  61#include "sdma.h"
  62
  63#undef pr_fmt
  64#define pr_fmt(fmt) DRIVER_NAME ": " fmt
  65
  66/*
  67 * The size has to be longer than this string, so we can append
  68 * board/chip information to it in the initialization code.
  69 */
  70const char ib_hfi1_version[] = HFI1_DRIVER_VERSION "\n";
  71
  72DEFINE_SPINLOCK(hfi1_devs_lock);
  73LIST_HEAD(hfi1_dev_list);
  74DEFINE_MUTEX(hfi1_mutex);       /* general driver use */
  75
  76unsigned int hfi1_max_mtu = HFI1_DEFAULT_MAX_MTU;
  77module_param_named(max_mtu, hfi1_max_mtu, uint, S_IRUGO);
  78MODULE_PARM_DESC(max_mtu, "Set max MTU bytes, default is " __stringify(
  79                 HFI1_DEFAULT_MAX_MTU));
  80
  81unsigned int hfi1_cu = 1;
  82module_param_named(cu, hfi1_cu, uint, S_IRUGO);
  83MODULE_PARM_DESC(cu, "Credit return units");
  84
  85unsigned long hfi1_cap_mask = HFI1_CAP_MASK_DEFAULT;
  86static int hfi1_caps_set(const char *, const struct kernel_param *);
  87static int hfi1_caps_get(char *, const struct kernel_param *);
  88static const struct kernel_param_ops cap_ops = {
  89        .set = hfi1_caps_set,
  90        .get = hfi1_caps_get
  91};
  92module_param_cb(cap_mask, &cap_ops, &hfi1_cap_mask, S_IWUSR | S_IRUGO);
  93MODULE_PARM_DESC(cap_mask, "Bit mask of enabled/disabled HW features");
  94
  95MODULE_LICENSE("Dual BSD/GPL");
  96MODULE_DESCRIPTION("Intel Omni-Path Architecture driver");
  97MODULE_VERSION(HFI1_DRIVER_VERSION);
  98
  99/*
 100 * MAX_PKT_RCV is the max # if packets processed per receive interrupt.
 101 */
 102#define MAX_PKT_RECV 64
 103#define EGR_HEAD_UPDATE_THRESHOLD 16
 104
 105struct hfi1_ib_stats hfi1_stats;
 106
 107static int hfi1_caps_set(const char *val, const struct kernel_param *kp)
 108{
 109        int ret = 0;
 110        unsigned long *cap_mask_ptr = (unsigned long *)kp->arg,
 111                cap_mask = *cap_mask_ptr, value, diff,
 112                write_mask = ((HFI1_CAP_WRITABLE_MASK << HFI1_CAP_USER_SHIFT) |
 113                              HFI1_CAP_WRITABLE_MASK);
 114
 115        ret = kstrtoul(val, 0, &value);
 116        if (ret) {
 117                pr_warn("Invalid module parameter value for 'cap_mask'\n");
 118                goto done;
 119        }
 120        /* Get the changed bits (except the locked bit) */
 121        diff = value ^ (cap_mask & ~HFI1_CAP_LOCKED_SMASK);
 122
 123        /* Remove any bits that are not allowed to change after driver load */
 124        if (HFI1_CAP_LOCKED() && (diff & ~write_mask)) {
 125                pr_warn("Ignoring non-writable capability bits %#lx\n",
 126                        diff & ~write_mask);
 127                diff &= write_mask;
 128        }
 129
 130        /* Mask off any reserved bits */
 131        diff &= ~HFI1_CAP_RESERVED_MASK;
 132        /* Clear any previously set and changing bits */
 133        cap_mask &= ~diff;
 134        /* Update the bits with the new capability */
 135        cap_mask |= (value & diff);
 136        /* Check for any kernel/user restrictions */
 137        diff = (cap_mask & (HFI1_CAP_MUST_HAVE_KERN << HFI1_CAP_USER_SHIFT)) ^
 138                ((cap_mask & HFI1_CAP_MUST_HAVE_KERN) << HFI1_CAP_USER_SHIFT);
 139        cap_mask &= ~diff;
 140        /* Set the bitmask to the final set */
 141        *cap_mask_ptr = cap_mask;
 142done:
 143        return ret;
 144}
 145
 146static int hfi1_caps_get(char *buffer, const struct kernel_param *kp)
 147{
 148        unsigned long cap_mask = *(unsigned long *)kp->arg;
 149
 150        cap_mask &= ~HFI1_CAP_LOCKED_SMASK;
 151        cap_mask |= ((cap_mask & HFI1_CAP_K2U) << HFI1_CAP_USER_SHIFT);
 152
 153        return scnprintf(buffer, PAGE_SIZE, "0x%lx", cap_mask);
 154}
 155
 156const char *get_unit_name(int unit)
 157{
 158        static char iname[16];
 159
 160        snprintf(iname, sizeof(iname), DRIVER_NAME "_%u", unit);
 161        return iname;
 162}
 163
 164const char *get_card_name(struct rvt_dev_info *rdi)
 165{
 166        struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
 167        struct hfi1_devdata *dd = container_of(ibdev,
 168                                               struct hfi1_devdata, verbs_dev);
 169        return get_unit_name(dd->unit);
 170}
 171
 172struct pci_dev *get_pci_dev(struct rvt_dev_info *rdi)
 173{
 174        struct hfi1_ibdev *ibdev = container_of(rdi, struct hfi1_ibdev, rdi);
 175        struct hfi1_devdata *dd = container_of(ibdev,
 176                                               struct hfi1_devdata, verbs_dev);
 177        return dd->pcidev;
 178}
 179
 180/*
 181 * Return count of units with at least one port ACTIVE.
 182 */
 183int hfi1_count_active_units(void)
 184{
 185        struct hfi1_devdata *dd;
 186        struct hfi1_pportdata *ppd;
 187        unsigned long flags;
 188        int pidx, nunits_active = 0;
 189
 190        spin_lock_irqsave(&hfi1_devs_lock, flags);
 191        list_for_each_entry(dd, &hfi1_dev_list, list) {
 192                if (!(dd->flags & HFI1_PRESENT) || !dd->kregbase)
 193                        continue;
 194                for (pidx = 0; pidx < dd->num_pports; ++pidx) {
 195                        ppd = dd->pport + pidx;
 196                        if (ppd->lid && ppd->linkup) {
 197                                nunits_active++;
 198                                break;
 199                        }
 200                }
 201        }
 202        spin_unlock_irqrestore(&hfi1_devs_lock, flags);
 203        return nunits_active;
 204}
 205
 206/*
 207 * Return count of all units, optionally return in arguments
 208 * the number of usable (present) units, and the number of
 209 * ports that are up.
 210 */
 211int hfi1_count_units(int *npresentp, int *nupp)
 212{
 213        int nunits = 0, npresent = 0, nup = 0;
 214        struct hfi1_devdata *dd;
 215        unsigned long flags;
 216        int pidx;
 217        struct hfi1_pportdata *ppd;
 218
 219        spin_lock_irqsave(&hfi1_devs_lock, flags);
 220
 221        list_for_each_entry(dd, &hfi1_dev_list, list) {
 222                nunits++;
 223                if ((dd->flags & HFI1_PRESENT) && dd->kregbase)
 224                        npresent++;
 225                for (pidx = 0; pidx < dd->num_pports; ++pidx) {
 226                        ppd = dd->pport + pidx;
 227                        if (ppd->lid && ppd->linkup)
 228                                nup++;
 229                }
 230        }
 231
 232        spin_unlock_irqrestore(&hfi1_devs_lock, flags);
 233
 234        if (npresentp)
 235                *npresentp = npresent;
 236        if (nupp)
 237                *nupp = nup;
 238
 239        return nunits;
 240}
 241
 242/*
 243 * Get address of eager buffer from it's index (allocated in chunks, not
 244 * contiguous).
 245 */
 246static inline void *get_egrbuf(const struct hfi1_ctxtdata *rcd, u64 rhf,
 247                               u8 *update)
 248{
 249        u32 idx = rhf_egr_index(rhf), offset = rhf_egr_buf_offset(rhf);
 250
 251        *update |= !(idx & (rcd->egrbufs.threshold - 1)) && !offset;
 252        return (void *)(((u64)(rcd->egrbufs.rcvtids[idx].addr)) +
 253                        (offset * RCV_BUF_BLOCK_SIZE));
 254}
 255
 256/*
 257 * Validate and encode the a given RcvArray Buffer size.
 258 * The function will check whether the given size falls within
 259 * allowed size ranges for the respective type and, optionally,
 260 * return the proper encoding.
 261 */
 262inline int hfi1_rcvbuf_validate(u32 size, u8 type, u16 *encoded)
 263{
 264        if (unlikely(!PAGE_ALIGNED(size)))
 265                return 0;
 266        if (unlikely(size < MIN_EAGER_BUFFER))
 267                return 0;
 268        if (size >
 269            (type == PT_EAGER ? MAX_EAGER_BUFFER : MAX_EXPECTED_BUFFER))
 270                return 0;
 271        if (encoded)
 272                *encoded = ilog2(size / PAGE_SIZE) + 1;
 273        return 1;
 274}
 275
 276static void rcv_hdrerr(struct hfi1_ctxtdata *rcd, struct hfi1_pportdata *ppd,
 277                       struct hfi1_packet *packet)
 278{
 279        struct hfi1_message_header *rhdr = packet->hdr;
 280        u32 rte = rhf_rcv_type_err(packet->rhf);
 281        int lnh = be16_to_cpu(rhdr->lrh[0]) & 3;
 282        struct hfi1_ibport *ibp = &ppd->ibport_data;
 283        struct hfi1_devdata *dd = ppd->dd;
 284        struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
 285
 286        if (packet->rhf & (RHF_VCRC_ERR | RHF_ICRC_ERR))
 287                return;
 288
 289        if (packet->rhf & RHF_TID_ERR) {
 290                /* For TIDERR and RC QPs preemptively schedule a NAK */
 291                struct hfi1_ib_header *hdr = (struct hfi1_ib_header *)rhdr;
 292                struct hfi1_other_headers *ohdr = NULL;
 293                u32 tlen = rhf_pkt_len(packet->rhf); /* in bytes */
 294                u16 lid  = be16_to_cpu(hdr->lrh[1]);
 295                u32 qp_num;
 296                u32 rcv_flags = 0;
 297
 298                /* Sanity check packet */
 299                if (tlen < 24)
 300                        goto drop;
 301
 302                /* Check for GRH */
 303                if (lnh == HFI1_LRH_BTH) {
 304                        ohdr = &hdr->u.oth;
 305                } else if (lnh == HFI1_LRH_GRH) {
 306                        u32 vtf;
 307
 308                        ohdr = &hdr->u.l.oth;
 309                        if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
 310                                goto drop;
 311                        vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
 312                        if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
 313                                goto drop;
 314                        rcv_flags |= HFI1_HAS_GRH;
 315                } else {
 316                        goto drop;
 317                }
 318                /* Get the destination QP number. */
 319                qp_num = be32_to_cpu(ohdr->bth[1]) & RVT_QPN_MASK;
 320                if (lid < be16_to_cpu(IB_MULTICAST_LID_BASE)) {
 321                        struct rvt_qp *qp;
 322                        unsigned long flags;
 323
 324                        rcu_read_lock();
 325                        qp = rvt_lookup_qpn(rdi, &ibp->rvp, qp_num);
 326                        if (!qp) {
 327                                rcu_read_unlock();
 328                                goto drop;
 329                        }
 330
 331                        /*
 332                         * Handle only RC QPs - for other QP types drop error
 333                         * packet.
 334                         */
 335                        spin_lock_irqsave(&qp->r_lock, flags);
 336
 337                        /* Check for valid receive state. */
 338                        if (!(ib_rvt_state_ops[qp->state] &
 339                              RVT_PROCESS_RECV_OK)) {
 340                                ibp->rvp.n_pkt_drops++;
 341                        }
 342
 343                        switch (qp->ibqp.qp_type) {
 344                        case IB_QPT_RC:
 345                                hfi1_rc_hdrerr(
 346                                        rcd,
 347                                        hdr,
 348                                        rcv_flags,
 349                                        qp);
 350                                break;
 351                        default:
 352                                /* For now don't handle any other QP types */
 353                                break;
 354                        }
 355
 356                        spin_unlock_irqrestore(&qp->r_lock, flags);
 357                        rcu_read_unlock();
 358                } /* Unicast QP */
 359        } /* Valid packet with TIDErr */
 360
 361        /* handle "RcvTypeErr" flags */
 362        switch (rte) {
 363        case RHF_RTE_ERROR_OP_CODE_ERR:
 364        {
 365                u32 opcode;
 366                void *ebuf = NULL;
 367                __be32 *bth = NULL;
 368
 369                if (rhf_use_egr_bfr(packet->rhf))
 370                        ebuf = packet->ebuf;
 371
 372                if (!ebuf)
 373                        goto drop; /* this should never happen */
 374
 375                if (lnh == HFI1_LRH_BTH)
 376                        bth = (__be32 *)ebuf;
 377                else if (lnh == HFI1_LRH_GRH)
 378                        bth = (__be32 *)((char *)ebuf + sizeof(struct ib_grh));
 379                else
 380                        goto drop;
 381
 382                opcode = be32_to_cpu(bth[0]) >> 24;
 383                opcode &= 0xff;
 384
 385                if (opcode == IB_OPCODE_CNP) {
 386                        /*
 387                         * Only in pre-B0 h/w is the CNP_OPCODE handled
 388                         * via this code path.
 389                         */
 390                        struct rvt_qp *qp = NULL;
 391                        u32 lqpn, rqpn;
 392                        u16 rlid;
 393                        u8 svc_type, sl, sc5;
 394
 395                        sc5  = (be16_to_cpu(rhdr->lrh[0]) >> 12) & 0xf;
 396                        if (rhf_dc_info(packet->rhf))
 397                                sc5 |= 0x10;
 398                        sl = ibp->sc_to_sl[sc5];
 399
 400                        lqpn = be32_to_cpu(bth[1]) & RVT_QPN_MASK;
 401                        rcu_read_lock();
 402                        qp = rvt_lookup_qpn(rdi, &ibp->rvp, lqpn);
 403                        if (!qp) {
 404                                rcu_read_unlock();
 405                                goto drop;
 406                        }
 407
 408                        switch (qp->ibqp.qp_type) {
 409                        case IB_QPT_UD:
 410                                rlid = 0;
 411                                rqpn = 0;
 412                                svc_type = IB_CC_SVCTYPE_UD;
 413                                break;
 414                        case IB_QPT_UC:
 415                                rlid = be16_to_cpu(rhdr->lrh[3]);
 416                                rqpn = qp->remote_qpn;
 417                                svc_type = IB_CC_SVCTYPE_UC;
 418                                break;
 419                        default:
 420                                goto drop;
 421                        }
 422
 423                        process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
 424                        rcu_read_unlock();
 425                }
 426
 427                packet->rhf &= ~RHF_RCV_TYPE_ERR_SMASK;
 428                break;
 429        }
 430        default:
 431                break;
 432        }
 433
 434drop:
 435        return;
 436}
 437
 438static inline void init_packet(struct hfi1_ctxtdata *rcd,
 439                               struct hfi1_packet *packet)
 440{
 441        packet->rsize = rcd->rcvhdrqentsize; /* words */
 442        packet->maxcnt = rcd->rcvhdrq_cnt * packet->rsize; /* words */
 443        packet->rcd = rcd;
 444        packet->updegr = 0;
 445        packet->etail = -1;
 446        packet->rhf_addr = get_rhf_addr(rcd);
 447        packet->rhf = rhf_to_cpu(packet->rhf_addr);
 448        packet->rhqoff = rcd->head;
 449        packet->numpkt = 0;
 450        packet->rcv_flags = 0;
 451}
 452
 453static void process_ecn(struct rvt_qp *qp, struct hfi1_ib_header *hdr,
 454                        struct hfi1_other_headers *ohdr,
 455                        u64 rhf, u32 bth1, struct ib_grh *grh)
 456{
 457        struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
 458        u32 rqpn = 0;
 459        u16 rlid;
 460        u8 sc5, svc_type;
 461
 462        switch (qp->ibqp.qp_type) {
 463        case IB_QPT_SMI:
 464        case IB_QPT_GSI:
 465        case IB_QPT_UD:
 466                rlid = be16_to_cpu(hdr->lrh[3]);
 467                rqpn = be32_to_cpu(ohdr->u.ud.deth[1]) & RVT_QPN_MASK;
 468                svc_type = IB_CC_SVCTYPE_UD;
 469                break;
 470        case IB_QPT_UC:
 471                rlid = qp->remote_ah_attr.dlid;
 472                rqpn = qp->remote_qpn;
 473                svc_type = IB_CC_SVCTYPE_UC;
 474                break;
 475        case IB_QPT_RC:
 476                rlid = qp->remote_ah_attr.dlid;
 477                rqpn = qp->remote_qpn;
 478                svc_type = IB_CC_SVCTYPE_RC;
 479                break;
 480        default:
 481                return;
 482        }
 483
 484        sc5 = (be16_to_cpu(hdr->lrh[0]) >> 12) & 0xf;
 485        if (rhf_dc_info(rhf))
 486                sc5 |= 0x10;
 487
 488        if (bth1 & HFI1_FECN_SMASK) {
 489                u16 pkey = (u16)be32_to_cpu(ohdr->bth[0]);
 490                u16 dlid = be16_to_cpu(hdr->lrh[1]);
 491
 492                return_cnp(ibp, qp, rqpn, pkey, dlid, rlid, sc5, grh);
 493        }
 494
 495        if (bth1 & HFI1_BECN_SMASK) {
 496                struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
 497                u32 lqpn = bth1 & RVT_QPN_MASK;
 498                u8 sl = ibp->sc_to_sl[sc5];
 499
 500                process_becn(ppd, sl, rlid, lqpn, rqpn, svc_type);
 501        }
 502}
 503
 504struct ps_mdata {
 505        struct hfi1_ctxtdata *rcd;
 506        u32 rsize;
 507        u32 maxcnt;
 508        u32 ps_head;
 509        u32 ps_tail;
 510        u32 ps_seq;
 511};
 512
 513static inline void init_ps_mdata(struct ps_mdata *mdata,
 514                                 struct hfi1_packet *packet)
 515{
 516        struct hfi1_ctxtdata *rcd = packet->rcd;
 517
 518        mdata->rcd = rcd;
 519        mdata->rsize = packet->rsize;
 520        mdata->maxcnt = packet->maxcnt;
 521        mdata->ps_head = packet->rhqoff;
 522
 523        if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
 524                mdata->ps_tail = get_rcvhdrtail(rcd);
 525                if (rcd->ctxt == HFI1_CTRL_CTXT)
 526                        mdata->ps_seq = rcd->seq_cnt;
 527                else
 528                        mdata->ps_seq = 0; /* not used with DMA_RTAIL */
 529        } else {
 530                mdata->ps_tail = 0; /* used only with DMA_RTAIL*/
 531                mdata->ps_seq = rcd->seq_cnt;
 532        }
 533}
 534
 535static inline int ps_done(struct ps_mdata *mdata, u64 rhf,
 536                          struct hfi1_ctxtdata *rcd)
 537{
 538        if (HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL))
 539                return mdata->ps_head == mdata->ps_tail;
 540        return mdata->ps_seq != rhf_rcv_seq(rhf);
 541}
 542
 543static inline int ps_skip(struct ps_mdata *mdata, u64 rhf,
 544                          struct hfi1_ctxtdata *rcd)
 545{
 546        /*
 547         * Control context can potentially receive an invalid rhf.
 548         * Drop such packets.
 549         */
 550        if ((rcd->ctxt == HFI1_CTRL_CTXT) && (mdata->ps_head != mdata->ps_tail))
 551                return mdata->ps_seq != rhf_rcv_seq(rhf);
 552
 553        return 0;
 554}
 555
 556static inline void update_ps_mdata(struct ps_mdata *mdata,
 557                                   struct hfi1_ctxtdata *rcd)
 558{
 559        mdata->ps_head += mdata->rsize;
 560        if (mdata->ps_head >= mdata->maxcnt)
 561                mdata->ps_head = 0;
 562
 563        /* Control context must do seq counting */
 564        if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL) ||
 565            (rcd->ctxt == HFI1_CTRL_CTXT)) {
 566                if (++mdata->ps_seq > 13)
 567                        mdata->ps_seq = 1;
 568        }
 569}
 570
 571/*
 572 * prescan_rxq - search through the receive queue looking for packets
 573 * containing Excplicit Congestion Notifications (FECNs, or BECNs).
 574 * When an ECN is found, process the Congestion Notification, and toggle
 575 * it off.
 576 * This is declared as a macro to allow quick checking of the port to avoid
 577 * the overhead of a function call if not enabled.
 578 */
 579#define prescan_rxq(rcd, packet) \
 580        do { \
 581                if (rcd->ppd->cc_prescan) \
 582                        __prescan_rxq(packet); \
 583        } while (0)
 584static void __prescan_rxq(struct hfi1_packet *packet)
 585{
 586        struct hfi1_ctxtdata *rcd = packet->rcd;
 587        struct ps_mdata mdata;
 588
 589        init_ps_mdata(&mdata, packet);
 590
 591        while (1) {
 592                struct hfi1_devdata *dd = rcd->dd;
 593                struct hfi1_ibport *ibp = &rcd->ppd->ibport_data;
 594                __le32 *rhf_addr = (__le32 *)rcd->rcvhdrq + mdata.ps_head +
 595                                         dd->rhf_offset;
 596                struct rvt_qp *qp;
 597                struct hfi1_ib_header *hdr;
 598                struct hfi1_other_headers *ohdr;
 599                struct ib_grh *grh = NULL;
 600                struct rvt_dev_info *rdi = &dd->verbs_dev.rdi;
 601                u64 rhf = rhf_to_cpu(rhf_addr);
 602                u32 etype = rhf_rcv_type(rhf), qpn, bth1;
 603                int is_ecn = 0;
 604                u8 lnh;
 605
 606                if (ps_done(&mdata, rhf, rcd))
 607                        break;
 608
 609                if (ps_skip(&mdata, rhf, rcd))
 610                        goto next;
 611
 612                if (etype != RHF_RCV_TYPE_IB)
 613                        goto next;
 614
 615                hdr = (struct hfi1_ib_header *)
 616                        hfi1_get_msgheader(dd, rhf_addr);
 617                lnh = be16_to_cpu(hdr->lrh[0]) & 3;
 618
 619                if (lnh == HFI1_LRH_BTH) {
 620                        ohdr = &hdr->u.oth;
 621                } else if (lnh == HFI1_LRH_GRH) {
 622                        ohdr = &hdr->u.l.oth;
 623                        grh = &hdr->u.l.grh;
 624                } else {
 625                        goto next; /* just in case */
 626                }
 627                bth1 = be32_to_cpu(ohdr->bth[1]);
 628                is_ecn = !!(bth1 & (HFI1_FECN_SMASK | HFI1_BECN_SMASK));
 629
 630                if (!is_ecn)
 631                        goto next;
 632
 633                qpn = bth1 & RVT_QPN_MASK;
 634                rcu_read_lock();
 635                qp = rvt_lookup_qpn(rdi, &ibp->rvp, qpn);
 636
 637                if (!qp) {
 638                        rcu_read_unlock();
 639                        goto next;
 640                }
 641
 642                process_ecn(qp, hdr, ohdr, rhf, bth1, grh);
 643                rcu_read_unlock();
 644
 645                /* turn off BECN, FECN */
 646                bth1 &= ~(HFI1_FECN_SMASK | HFI1_BECN_SMASK);
 647                ohdr->bth[1] = cpu_to_be32(bth1);
 648next:
 649                update_ps_mdata(&mdata, rcd);
 650        }
 651}
 652
 653static inline int skip_rcv_packet(struct hfi1_packet *packet, int thread)
 654{
 655        int ret = RCV_PKT_OK;
 656
 657        /* Set up for the next packet */
 658        packet->rhqoff += packet->rsize;
 659        if (packet->rhqoff >= packet->maxcnt)
 660                packet->rhqoff = 0;
 661
 662        packet->numpkt++;
 663        if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) {
 664                if (thread) {
 665                        cond_resched();
 666                } else {
 667                        ret = RCV_PKT_LIMIT;
 668                        this_cpu_inc(*packet->rcd->dd->rcv_limit);
 669                }
 670        }
 671
 672        packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
 673                                     packet->rcd->dd->rhf_offset;
 674        packet->rhf = rhf_to_cpu(packet->rhf_addr);
 675
 676        return ret;
 677}
 678
 679static inline int process_rcv_packet(struct hfi1_packet *packet, int thread)
 680{
 681        int ret = RCV_PKT_OK;
 682
 683        packet->hdr = hfi1_get_msgheader(packet->rcd->dd,
 684                                         packet->rhf_addr);
 685        packet->hlen = (u8 *)packet->rhf_addr - (u8 *)packet->hdr;
 686        packet->etype = rhf_rcv_type(packet->rhf);
 687        /* total length */
 688        packet->tlen = rhf_pkt_len(packet->rhf); /* in bytes */
 689        /* retrieve eager buffer details */
 690        packet->ebuf = NULL;
 691        if (rhf_use_egr_bfr(packet->rhf)) {
 692                packet->etail = rhf_egr_index(packet->rhf);
 693                packet->ebuf = get_egrbuf(packet->rcd, packet->rhf,
 694                                 &packet->updegr);
 695                /*
 696                 * Prefetch the contents of the eager buffer.  It is
 697                 * OK to send a negative length to prefetch_range().
 698                 * The +2 is the size of the RHF.
 699                 */
 700                prefetch_range(packet->ebuf,
 701                               packet->tlen - ((packet->rcd->rcvhdrqentsize -
 702                                               (rhf_hdrq_offset(packet->rhf)
 703                                                + 2)) * 4));
 704        }
 705
 706        /*
 707         * Call a type specific handler for the packet. We
 708         * should be able to trust that etype won't be beyond
 709         * the range of valid indexes. If so something is really
 710         * wrong and we can probably just let things come
 711         * crashing down. There is no need to eat another
 712         * comparison in this performance critical code.
 713         */
 714        packet->rcd->dd->rhf_rcv_function_map[packet->etype](packet);
 715        packet->numpkt++;
 716
 717        /* Set up for the next packet */
 718        packet->rhqoff += packet->rsize;
 719        if (packet->rhqoff >= packet->maxcnt)
 720                packet->rhqoff = 0;
 721
 722        if (unlikely((packet->numpkt & (MAX_PKT_RECV - 1)) == 0)) {
 723                if (thread) {
 724                        cond_resched();
 725                } else {
 726                        ret = RCV_PKT_LIMIT;
 727                        this_cpu_inc(*packet->rcd->dd->rcv_limit);
 728                }
 729        }
 730
 731        packet->rhf_addr = (__le32 *)packet->rcd->rcvhdrq + packet->rhqoff +
 732                                      packet->rcd->dd->rhf_offset;
 733        packet->rhf = rhf_to_cpu(packet->rhf_addr);
 734
 735        return ret;
 736}
 737
 738static inline void process_rcv_update(int last, struct hfi1_packet *packet)
 739{
 740        /*
 741         * Update head regs etc., every 16 packets, if not last pkt,
 742         * to help prevent rcvhdrq overflows, when many packets
 743         * are processed and queue is nearly full.
 744         * Don't request an interrupt for intermediate updates.
 745         */
 746        if (!last && !(packet->numpkt & 0xf)) {
 747                update_usrhead(packet->rcd, packet->rhqoff, packet->updegr,
 748                               packet->etail, 0, 0);
 749                packet->updegr = 0;
 750        }
 751        packet->rcv_flags = 0;
 752}
 753
 754static inline void finish_packet(struct hfi1_packet *packet)
 755{
 756        /*
 757         * Nothing we need to free for the packet.
 758         *
 759         * The only thing we need to do is a final update and call for an
 760         * interrupt
 761         */
 762        update_usrhead(packet->rcd, packet->rcd->head, packet->updegr,
 763                       packet->etail, rcv_intr_dynamic, packet->numpkt);
 764}
 765
 766static inline void process_rcv_qp_work(struct hfi1_packet *packet)
 767{
 768        struct hfi1_ctxtdata *rcd;
 769        struct rvt_qp *qp, *nqp;
 770
 771        rcd = packet->rcd;
 772        rcd->head = packet->rhqoff;
 773
 774        /*
 775         * Iterate over all QPs waiting to respond.
 776         * The list won't change since the IRQ is only run on one CPU.
 777         */
 778        list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
 779                list_del_init(&qp->rspwait);
 780                if (qp->r_flags & RVT_R_RSP_NAK) {
 781                        qp->r_flags &= ~RVT_R_RSP_NAK;
 782                        hfi1_send_rc_ack(rcd, qp, 0);
 783                }
 784                if (qp->r_flags & RVT_R_RSP_SEND) {
 785                        unsigned long flags;
 786
 787                        qp->r_flags &= ~RVT_R_RSP_SEND;
 788                        spin_lock_irqsave(&qp->s_lock, flags);
 789                        if (ib_rvt_state_ops[qp->state] &
 790                                        RVT_PROCESS_OR_FLUSH_SEND)
 791                                hfi1_schedule_send(qp);
 792                        spin_unlock_irqrestore(&qp->s_lock, flags);
 793                }
 794                if (atomic_dec_and_test(&qp->refcount))
 795                        wake_up(&qp->wait);
 796        }
 797}
 798
 799/*
 800 * Handle receive interrupts when using the no dma rtail option.
 801 */
 802int handle_receive_interrupt_nodma_rtail(struct hfi1_ctxtdata *rcd, int thread)
 803{
 804        u32 seq;
 805        int last = RCV_PKT_OK;
 806        struct hfi1_packet packet;
 807
 808        init_packet(rcd, &packet);
 809        seq = rhf_rcv_seq(packet.rhf);
 810        if (seq != rcd->seq_cnt) {
 811                last = RCV_PKT_DONE;
 812                goto bail;
 813        }
 814
 815        prescan_rxq(rcd, &packet);
 816
 817        while (last == RCV_PKT_OK) {
 818                last = process_rcv_packet(&packet, thread);
 819                seq = rhf_rcv_seq(packet.rhf);
 820                if (++rcd->seq_cnt > 13)
 821                        rcd->seq_cnt = 1;
 822                if (seq != rcd->seq_cnt)
 823                        last = RCV_PKT_DONE;
 824                process_rcv_update(last, &packet);
 825        }
 826        process_rcv_qp_work(&packet);
 827bail:
 828        finish_packet(&packet);
 829        return last;
 830}
 831
 832int handle_receive_interrupt_dma_rtail(struct hfi1_ctxtdata *rcd, int thread)
 833{
 834        u32 hdrqtail;
 835        int last = RCV_PKT_OK;
 836        struct hfi1_packet packet;
 837
 838        init_packet(rcd, &packet);
 839        hdrqtail = get_rcvhdrtail(rcd);
 840        if (packet.rhqoff == hdrqtail) {
 841                last = RCV_PKT_DONE;
 842                goto bail;
 843        }
 844        smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
 845
 846        prescan_rxq(rcd, &packet);
 847
 848        while (last == RCV_PKT_OK) {
 849                last = process_rcv_packet(&packet, thread);
 850                if (packet.rhqoff == hdrqtail)
 851                        last = RCV_PKT_DONE;
 852                process_rcv_update(last, &packet);
 853        }
 854        process_rcv_qp_work(&packet);
 855bail:
 856        finish_packet(&packet);
 857        return last;
 858}
 859
 860static inline void set_all_nodma_rtail(struct hfi1_devdata *dd)
 861{
 862        int i;
 863
 864        for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
 865                dd->rcd[i]->do_interrupt =
 866                        &handle_receive_interrupt_nodma_rtail;
 867}
 868
 869static inline void set_all_dma_rtail(struct hfi1_devdata *dd)
 870{
 871        int i;
 872
 873        for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
 874                dd->rcd[i]->do_interrupt =
 875                        &handle_receive_interrupt_dma_rtail;
 876}
 877
 878void set_all_slowpath(struct hfi1_devdata *dd)
 879{
 880        int i;
 881
 882        /* HFI1_CTRL_CTXT must always use the slow path interrupt handler */
 883        for (i = HFI1_CTRL_CTXT + 1; i < dd->first_user_ctxt; i++)
 884                dd->rcd[i]->do_interrupt = &handle_receive_interrupt;
 885}
 886
 887static inline int set_armed_to_active(struct hfi1_ctxtdata *rcd,
 888                                      struct hfi1_packet packet,
 889                                      struct hfi1_devdata *dd)
 890{
 891        struct work_struct *lsaw = &rcd->ppd->linkstate_active_work;
 892        struct hfi1_message_header *hdr = hfi1_get_msgheader(packet.rcd->dd,
 893                                                             packet.rhf_addr);
 894
 895        if (hdr2sc(hdr, packet.rhf) != 0xf) {
 896                int hwstate = read_logical_state(dd);
 897
 898                if (hwstate != LSTATE_ACTIVE) {
 899                        dd_dev_info(dd, "Unexpected link state %d\n", hwstate);
 900                        return 0;
 901                }
 902
 903                queue_work(rcd->ppd->hfi1_wq, lsaw);
 904                return 1;
 905        }
 906        return 0;
 907}
 908
 909/*
 910 * handle_receive_interrupt - receive a packet
 911 * @rcd: the context
 912 *
 913 * Called from interrupt handler for errors or receive interrupt.
 914 * This is the slow path interrupt handler.
 915 */
 916int handle_receive_interrupt(struct hfi1_ctxtdata *rcd, int thread)
 917{
 918        struct hfi1_devdata *dd = rcd->dd;
 919        u32 hdrqtail;
 920        int needset, last = RCV_PKT_OK;
 921        struct hfi1_packet packet;
 922        int skip_pkt = 0;
 923
 924        /* Control context will always use the slow path interrupt handler */
 925        needset = (rcd->ctxt == HFI1_CTRL_CTXT) ? 0 : 1;
 926
 927        init_packet(rcd, &packet);
 928
 929        if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
 930                u32 seq = rhf_rcv_seq(packet.rhf);
 931
 932                if (seq != rcd->seq_cnt) {
 933                        last = RCV_PKT_DONE;
 934                        goto bail;
 935                }
 936                hdrqtail = 0;
 937        } else {
 938                hdrqtail = get_rcvhdrtail(rcd);
 939                if (packet.rhqoff == hdrqtail) {
 940                        last = RCV_PKT_DONE;
 941                        goto bail;
 942                }
 943                smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
 944
 945                /*
 946                 * Control context can potentially receive an invalid
 947                 * rhf. Drop such packets.
 948                 */
 949                if (rcd->ctxt == HFI1_CTRL_CTXT) {
 950                        u32 seq = rhf_rcv_seq(packet.rhf);
 951
 952                        if (seq != rcd->seq_cnt)
 953                                skip_pkt = 1;
 954                }
 955        }
 956
 957        prescan_rxq(rcd, &packet);
 958
 959        while (last == RCV_PKT_OK) {
 960                if (unlikely(dd->do_drop &&
 961                             atomic_xchg(&dd->drop_packet, DROP_PACKET_OFF) ==
 962                             DROP_PACKET_ON)) {
 963                        dd->do_drop = 0;
 964
 965                        /* On to the next packet */
 966                        packet.rhqoff += packet.rsize;
 967                        packet.rhf_addr = (__le32 *)rcd->rcvhdrq +
 968                                          packet.rhqoff +
 969                                          dd->rhf_offset;
 970                        packet.rhf = rhf_to_cpu(packet.rhf_addr);
 971
 972                } else if (skip_pkt) {
 973                        last = skip_rcv_packet(&packet, thread);
 974                        skip_pkt = 0;
 975                } else {
 976                        /* Auto activate link on non-SC15 packet receive */
 977                        if (unlikely(rcd->ppd->host_link_state ==
 978                                     HLS_UP_ARMED) &&
 979                            set_armed_to_active(rcd, packet, dd))
 980                                goto bail;
 981                        last = process_rcv_packet(&packet, thread);
 982                }
 983
 984                if (!HFI1_CAP_KGET_MASK(rcd->flags, DMA_RTAIL)) {
 985                        u32 seq = rhf_rcv_seq(packet.rhf);
 986
 987                        if (++rcd->seq_cnt > 13)
 988                                rcd->seq_cnt = 1;
 989                        if (seq != rcd->seq_cnt)
 990                                last = RCV_PKT_DONE;
 991                        if (needset) {
 992                                dd_dev_info(dd, "Switching to NO_DMA_RTAIL\n");
 993                                set_all_nodma_rtail(dd);
 994                                needset = 0;
 995                        }
 996                } else {
 997                        if (packet.rhqoff == hdrqtail)
 998                                last = RCV_PKT_DONE;
 999                        /*
1000                         * Control context can potentially receive an invalid
1001                         * rhf. Drop such packets.
1002                         */
1003                        if (rcd->ctxt == HFI1_CTRL_CTXT) {
1004                                u32 seq = rhf_rcv_seq(packet.rhf);
1005
1006                                if (++rcd->seq_cnt > 13)
1007                                        rcd->seq_cnt = 1;
1008                                if (!last && (seq != rcd->seq_cnt))
1009                                        skip_pkt = 1;
1010                        }
1011
1012                        if (needset) {
1013                                dd_dev_info(dd,
1014                                            "Switching to DMA_RTAIL\n");
1015                                set_all_dma_rtail(dd);
1016                                needset = 0;
1017                        }
1018                }
1019
1020                process_rcv_update(last, &packet);
1021        }
1022
1023        process_rcv_qp_work(&packet);
1024
1025bail:
1026        /*
1027         * Always write head at end, and setup rcv interrupt, even
1028         * if no packets were processed.
1029         */
1030        finish_packet(&packet);
1031        return last;
1032}
1033
1034/*
1035 * We may discover in the interrupt that the hardware link state has
1036 * changed from ARMED to ACTIVE (due to the arrival of a non-SC15 packet),
1037 * and we need to update the driver's notion of the link state.  We cannot
1038 * run set_link_state from interrupt context, so we queue this function on
1039 * a workqueue.
1040 *
1041 * We delay the regular interrupt processing until after the state changes
1042 * so that the link will be in the correct state by the time any application
1043 * we wake up attempts to send a reply to any message it received.
1044 * (Subsequent receive interrupts may possibly force the wakeup before we
1045 * update the link state.)
1046 *
1047 * The rcd is freed in hfi1_free_ctxtdata after hfi1_postinit_cleanup invokes
1048 * dd->f_cleanup(dd) to disable the interrupt handler and flush workqueues,
1049 * so we're safe from use-after-free of the rcd.
1050 */
1051void receive_interrupt_work(struct work_struct *work)
1052{
1053        struct hfi1_pportdata *ppd = container_of(work, struct hfi1_pportdata,
1054                                                  linkstate_active_work);
1055        struct hfi1_devdata *dd = ppd->dd;
1056        int i;
1057
1058        /* Received non-SC15 packet implies neighbor_normal */
1059        ppd->neighbor_normal = 1;
1060        set_link_state(ppd, HLS_UP_ACTIVE);
1061
1062        /*
1063         * Interrupt all kernel contexts that could have had an
1064         * interrupt during auto activation.
1065         */
1066        for (i = HFI1_CTRL_CTXT; i < dd->first_user_ctxt; i++)
1067                force_recv_intr(dd->rcd[i]);
1068}
1069
1070/*
1071 * Convert a given MTU size to the on-wire MAD packet enumeration.
1072 * Return -1 if the size is invalid.
1073 */
1074int mtu_to_enum(u32 mtu, int default_if_bad)
1075{
1076        switch (mtu) {
1077        case     0: return OPA_MTU_0;
1078        case   256: return OPA_MTU_256;
1079        case   512: return OPA_MTU_512;
1080        case  1024: return OPA_MTU_1024;
1081        case  2048: return OPA_MTU_2048;
1082        case  4096: return OPA_MTU_4096;
1083        case  8192: return OPA_MTU_8192;
1084        case 10240: return OPA_MTU_10240;
1085        }
1086        return default_if_bad;
1087}
1088
1089u16 enum_to_mtu(int mtu)
1090{
1091        switch (mtu) {
1092        case OPA_MTU_0:     return 0;
1093        case OPA_MTU_256:   return 256;
1094        case OPA_MTU_512:   return 512;
1095        case OPA_MTU_1024:  return 1024;
1096        case OPA_MTU_2048:  return 2048;
1097        case OPA_MTU_4096:  return 4096;
1098        case OPA_MTU_8192:  return 8192;
1099        case OPA_MTU_10240: return 10240;
1100        default: return 0xffff;
1101        }
1102}
1103
1104/*
1105 * set_mtu - set the MTU
1106 * @ppd: the per port data
1107 *
1108 * We can handle "any" incoming size, the issue here is whether we
1109 * need to restrict our outgoing size.  We do not deal with what happens
1110 * to programs that are already running when the size changes.
1111 */
1112int set_mtu(struct hfi1_pportdata *ppd)
1113{
1114        struct hfi1_devdata *dd = ppd->dd;
1115        int i, drain, ret = 0, is_up = 0;
1116
1117        ppd->ibmtu = 0;
1118        for (i = 0; i < ppd->vls_supported; i++)
1119                if (ppd->ibmtu < dd->vld[i].mtu)
1120                        ppd->ibmtu = dd->vld[i].mtu;
1121        ppd->ibmaxlen = ppd->ibmtu + lrh_max_header_bytes(ppd->dd);
1122
1123        mutex_lock(&ppd->hls_lock);
1124        if (ppd->host_link_state == HLS_UP_INIT ||
1125            ppd->host_link_state == HLS_UP_ARMED ||
1126            ppd->host_link_state == HLS_UP_ACTIVE)
1127                is_up = 1;
1128
1129        drain = !is_ax(dd) && is_up;
1130
1131        if (drain)
1132                /*
1133                 * MTU is specified per-VL. To ensure that no packet gets
1134                 * stuck (due, e.g., to the MTU for the packet's VL being
1135                 * reduced), empty the per-VL FIFOs before adjusting MTU.
1136                 */
1137                ret = stop_drain_data_vls(dd);
1138
1139        if (ret) {
1140                dd_dev_err(dd, "%s: cannot stop/drain VLs - refusing to change per-VL MTUs\n",
1141                           __func__);
1142                goto err;
1143        }
1144
1145        hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_MTU, 0);
1146
1147        if (drain)
1148                open_fill_data_vls(dd); /* reopen all VLs */
1149
1150err:
1151        mutex_unlock(&ppd->hls_lock);
1152
1153        return ret;
1154}
1155
1156int hfi1_set_lid(struct hfi1_pportdata *ppd, u32 lid, u8 lmc)
1157{
1158        struct hfi1_devdata *dd = ppd->dd;
1159
1160        ppd->lid = lid;
1161        ppd->lmc = lmc;
1162        hfi1_set_ib_cfg(ppd, HFI1_IB_CFG_LIDLMC, 0);
1163
1164        dd_dev_info(dd, "port %u: got a lid: 0x%x\n", ppd->port, lid);
1165
1166        return 0;
1167}
1168
1169void shutdown_led_override(struct hfi1_pportdata *ppd)
1170{
1171        struct hfi1_devdata *dd = ppd->dd;
1172
1173        /*
1174         * This pairs with the memory barrier in hfi1_start_led_override to
1175         * ensure that we read the correct state of LED beaconing represented
1176         * by led_override_timer_active
1177         */
1178        smp_rmb();
1179        if (atomic_read(&ppd->led_override_timer_active)) {
1180                del_timer_sync(&ppd->led_override_timer);
1181                atomic_set(&ppd->led_override_timer_active, 0);
1182                /* Ensure the atomic_set is visible to all CPUs */
1183                smp_wmb();
1184        }
1185
1186        /* Hand control of the LED to the DC for normal operation */
1187        write_csr(dd, DCC_CFG_LED_CNTRL, 0);
1188}
1189
1190static void run_led_override(unsigned long opaque)
1191{
1192        struct hfi1_pportdata *ppd = (struct hfi1_pportdata *)opaque;
1193        struct hfi1_devdata *dd = ppd->dd;
1194        unsigned long timeout;
1195        int phase_idx;
1196
1197        if (!(dd->flags & HFI1_INITTED))
1198                return;
1199
1200        phase_idx = ppd->led_override_phase & 1;
1201
1202        setextled(dd, phase_idx);
1203
1204        timeout = ppd->led_override_vals[phase_idx];
1205
1206        /* Set up for next phase */
1207        ppd->led_override_phase = !ppd->led_override_phase;
1208
1209        mod_timer(&ppd->led_override_timer, jiffies + timeout);
1210}
1211
1212/*
1213 * To have the LED blink in a particular pattern, provide timeon and timeoff
1214 * in milliseconds.
1215 * To turn off custom blinking and return to normal operation, use
1216 * shutdown_led_override()
1217 */
1218void hfi1_start_led_override(struct hfi1_pportdata *ppd, unsigned int timeon,
1219                             unsigned int timeoff)
1220{
1221        if (!(ppd->dd->flags & HFI1_INITTED))
1222                return;
1223
1224        /* Convert to jiffies for direct use in timer */
1225        ppd->led_override_vals[0] = msecs_to_jiffies(timeoff);
1226        ppd->led_override_vals[1] = msecs_to_jiffies(timeon);
1227
1228        /* Arbitrarily start from LED on phase */
1229        ppd->led_override_phase = 1;
1230
1231        /*
1232         * If the timer has not already been started, do so. Use a "quick"
1233         * timeout so the handler will be called soon to look at our request.
1234         */
1235        if (!timer_pending(&ppd->led_override_timer)) {
1236                setup_timer(&ppd->led_override_timer, run_led_override,
1237                            (unsigned long)ppd);
1238                ppd->led_override_timer.expires = jiffies + 1;
1239                add_timer(&ppd->led_override_timer);
1240                atomic_set(&ppd->led_override_timer_active, 1);
1241                /* Ensure the atomic_set is visible to all CPUs */
1242                smp_wmb();
1243        }
1244}
1245
1246/**
1247 * hfi1_reset_device - reset the chip if possible
1248 * @unit: the device to reset
1249 *
1250 * Whether or not reset is successful, we attempt to re-initialize the chip
1251 * (that is, much like a driver unload/reload).  We clear the INITTED flag
1252 * so that the various entry points will fail until we reinitialize.  For
1253 * now, we only allow this if no user contexts are open that use chip resources
1254 */
1255int hfi1_reset_device(int unit)
1256{
1257        int ret, i;
1258        struct hfi1_devdata *dd = hfi1_lookup(unit);
1259        struct hfi1_pportdata *ppd;
1260        unsigned long flags;
1261        int pidx;
1262
1263        if (!dd) {
1264                ret = -ENODEV;
1265                goto bail;
1266        }
1267
1268        dd_dev_info(dd, "Reset on unit %u requested\n", unit);
1269
1270        if (!dd->kregbase || !(dd->flags & HFI1_PRESENT)) {
1271                dd_dev_info(dd,
1272                            "Invalid unit number %u or not initialized or not present\n",
1273                            unit);
1274                ret = -ENXIO;
1275                goto bail;
1276        }
1277
1278        spin_lock_irqsave(&dd->uctxt_lock, flags);
1279        if (dd->rcd)
1280                for (i = dd->first_user_ctxt; i < dd->num_rcv_contexts; i++) {
1281                        if (!dd->rcd[i] || !dd->rcd[i]->cnt)
1282                                continue;
1283                        spin_unlock_irqrestore(&dd->uctxt_lock, flags);
1284                        ret = -EBUSY;
1285                        goto bail;
1286                }
1287        spin_unlock_irqrestore(&dd->uctxt_lock, flags);
1288
1289        for (pidx = 0; pidx < dd->num_pports; ++pidx) {
1290                ppd = dd->pport + pidx;
1291
1292                shutdown_led_override(ppd);
1293        }
1294        if (dd->flags & HFI1_HAS_SEND_DMA)
1295                sdma_exit(dd);
1296
1297        hfi1_reset_cpu_counters(dd);
1298
1299        ret = hfi1_init(dd, 1);
1300
1301        if (ret)
1302                dd_dev_err(dd,
1303                           "Reinitialize unit %u after reset failed with %d\n",
1304                           unit, ret);
1305        else
1306                dd_dev_info(dd, "Reinitialized unit %u after resetting\n",
1307                            unit);
1308
1309bail:
1310        return ret;
1311}
1312
1313void handle_eflags(struct hfi1_packet *packet)
1314{
1315        struct hfi1_ctxtdata *rcd = packet->rcd;
1316        u32 rte = rhf_rcv_type_err(packet->rhf);
1317
1318        rcv_hdrerr(rcd, rcd->ppd, packet);
1319        if (rhf_err_flags(packet->rhf))
1320                dd_dev_err(rcd->dd,
1321                           "receive context %d: rhf 0x%016llx, errs [ %s%s%s%s%s%s%s%s] rte 0x%x\n",
1322                           rcd->ctxt, packet->rhf,
1323                           packet->rhf & RHF_K_HDR_LEN_ERR ? "k_hdr_len " : "",
1324                           packet->rhf & RHF_DC_UNC_ERR ? "dc_unc " : "",
1325                           packet->rhf & RHF_DC_ERR ? "dc " : "",
1326                           packet->rhf & RHF_TID_ERR ? "tid " : "",
1327                           packet->rhf & RHF_LEN_ERR ? "len " : "",
1328                           packet->rhf & RHF_ECC_ERR ? "ecc " : "",
1329                           packet->rhf & RHF_VCRC_ERR ? "vcrc " : "",
1330                           packet->rhf & RHF_ICRC_ERR ? "icrc " : "",
1331                           rte);
1332}
1333
1334/*
1335 * The following functions are called by the interrupt handler. They are type
1336 * specific handlers for each packet type.
1337 */
1338int process_receive_ib(struct hfi1_packet *packet)
1339{
1340        trace_hfi1_rcvhdr(packet->rcd->ppd->dd,
1341                          packet->rcd->ctxt,
1342                          rhf_err_flags(packet->rhf),
1343                          RHF_RCV_TYPE_IB,
1344                          packet->hlen,
1345                          packet->tlen,
1346                          packet->updegr,
1347                          rhf_egr_index(packet->rhf));
1348
1349        if (unlikely(rhf_err_flags(packet->rhf))) {
1350                handle_eflags(packet);
1351                return RHF_RCV_CONTINUE;
1352        }
1353
1354        hfi1_ib_rcv(packet);
1355        return RHF_RCV_CONTINUE;
1356}
1357
1358int process_receive_bypass(struct hfi1_packet *packet)
1359{
1360        if (unlikely(rhf_err_flags(packet->rhf)))
1361                handle_eflags(packet);
1362
1363        dd_dev_err(packet->rcd->dd,
1364                   "Bypass packets are not supported in normal operation. Dropping\n");
1365        return RHF_RCV_CONTINUE;
1366}
1367
1368int process_receive_error(struct hfi1_packet *packet)
1369{
1370        handle_eflags(packet);
1371
1372        if (unlikely(rhf_err_flags(packet->rhf)))
1373                dd_dev_err(packet->rcd->dd,
1374                           "Unhandled error packet received. Dropping.\n");
1375
1376        return RHF_RCV_CONTINUE;
1377}
1378
1379int kdeth_process_expected(struct hfi1_packet *packet)
1380{
1381        if (unlikely(rhf_err_flags(packet->rhf)))
1382                handle_eflags(packet);
1383
1384        dd_dev_err(packet->rcd->dd,
1385                   "Unhandled expected packet received. Dropping.\n");
1386        return RHF_RCV_CONTINUE;
1387}
1388
1389int kdeth_process_eager(struct hfi1_packet *packet)
1390{
1391        if (unlikely(rhf_err_flags(packet->rhf)))
1392                handle_eflags(packet);
1393
1394        dd_dev_err(packet->rcd->dd,
1395                   "Unhandled eager packet received. Dropping.\n");
1396        return RHF_RCV_CONTINUE;
1397}
1398
1399int process_receive_invalid(struct hfi1_packet *packet)
1400{
1401        dd_dev_err(packet->rcd->dd, "Invalid packet type %d. Dropping\n",
1402                   rhf_rcv_type(packet->rhf));
1403        return RHF_RCV_CONTINUE;
1404}
1405