drivers/infiniband/hw/ipath/ipath_stats.c

   1 /*
   2  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, 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
  33 #include <linux/pci.h>
  34
  35 #include "ipath_kernel.h"
  36
  37 struct infinipath_stats ipath_stats;
  38
  39 /**
  40  * ipath_snap_cntr - snapshot a chip counter
  41  * @dd: the infinipath device
  42  * @creg: the counter to snapshot
  43  *
  44  * called from add_timer and user counter read calls, to deal with
  45  * counters that wrap in "human time".  The words sent and received, and
  46  * the packets sent and received are all that we worry about.  For now,
  47  * at least, we don't worry about error counters, because if they wrap
  48  * that quickly, we probably don't care.  We may eventually just make this
  49  * handle all the counters.  word counters can wrap in about 20 seconds
  50  * of full bandwidth traffic, packet counters in a few hours.
  51  */
  52
  53 u64 ipath_snap_cntr(struct ipath_devdata *dd, ipath_creg creg)
  54 {
  55         u32 val, reg64 = 0;
  56         u64 val64;
  57         unsigned long t0, t1;
  58         u64 ret;
  59
  60         t0 = jiffies;
  61         /* If fast increment counters are only 32 bits, snapshot them,
  62          * and maintain them as 64bit values in the driver */
  63         if (!(dd->ipath_flags & IPATH_32BITCOUNTERS) &&
  64             (creg == dd->ipath_cregs->cr_wordsendcnt ||
  65              creg == dd->ipath_cregs->cr_wordrcvcnt ||
  66              creg == dd->ipath_cregs->cr_pktsendcnt ||
  67              creg == dd->ipath_cregs->cr_pktrcvcnt)) {
  68                 val64 = ipath_read_creg(dd, creg);
  69                 val = val64 == ~0ULL ? ~0U : 0;
  70                 reg64 = 1;
  71         } else                  /* val64 just to keep gcc quiet... */
  72                 val64 = val = ipath_read_creg32(dd, creg);
  73         /*
  74          * See if a second has passed.  This is just a way to detect things
  75          * that are quite broken.  Normally this should take just a few
  76          * cycles (the check is for long enough that we don't care if we get
  77          * pre-empted.)  An Opteron HT O read timeout is 4 seconds with
  78          * normal NB values
  79          */
  80         t1 = jiffies;
  81         if (time_before(t0 + HZ, t1) && val == -1) {
  82                 ipath_dev_err(dd, "Error!  Read counter 0x%x timed out\n",
  83                               creg);
  84                 ret = 0ULL;
  85                 goto bail;
  86         }
  87         if (reg64) {
  88                 ret = val64;
  89                 goto bail;
  90         }
  91
  92         if (creg == dd->ipath_cregs->cr_wordsendcnt) {
  93                 if (val != dd->ipath_lastsword) {
  94                         dd->ipath_sword += val - dd->ipath_lastsword;
  95                         dd->ipath_lastsword = val;
  96                 }
  97                 val64 = dd->ipath_sword;
  98         } else if (creg == dd->ipath_cregs->cr_wordrcvcnt) {
  99                 if (val != dd->ipath_lastrword) {
 100                         dd->ipath_rword += val - dd->ipath_lastrword;
 101                         dd->ipath_lastrword = val;
 102                 }
 103                 val64 = dd->ipath_rword;
 104         } else if (creg == dd->ipath_cregs->cr_pktsendcnt) {
 105                 if (val != dd->ipath_lastspkts) {
 106                         dd->ipath_spkts += val - dd->ipath_lastspkts;
 107                         dd->ipath_lastspkts = val;
 108                 }
 109                 val64 = dd->ipath_spkts;
 110         } else if (creg == dd->ipath_cregs->cr_pktrcvcnt) {
 111                 if (val != dd->ipath_lastrpkts) {
 112                         dd->ipath_rpkts += val - dd->ipath_lastrpkts;
 113                         dd->ipath_lastrpkts = val;
 114                 }
 115                 val64 = dd->ipath_rpkts;
 116         } else
 117                 val64 = (u64) val;
 118
 119         ret = val64;
 120
 121 bail:
 122         return ret;
 123 }
 124
 125 /**
 126  * ipath_qcheck - print delta of egrfull/hdrqfull errors for kernel ports
 127  * @dd: the infinipath device
 128  *
 129  * print the delta of egrfull/hdrqfull errors for kernel ports no more than
 130  * every 5 seconds.  User processes are printed at close, but kernel doesn't
 131  * close, so...  Separate routine so may call from other places someday, and
 132  * so function name when printed by _IPATH_INFO is meaningfull
 133  */
 134 static void ipath_qcheck(struct ipath_devdata *dd)
 135 {
 136         static u64 last_tot_hdrqfull;
 137         size_t blen = 0;
 138         char buf[128];
 139
 140         *buf = 0;
 141         if (dd->ipath_pd[0]->port_hdrqfull != dd->ipath_p0_hdrqfull) {
 142                 blen = snprintf(buf, sizeof buf, "port 0 hdrqfull %u",
 143                                 dd->ipath_pd[0]->port_hdrqfull -
 144                                 dd->ipath_p0_hdrqfull);
 145                 dd->ipath_p0_hdrqfull = dd->ipath_pd[0]->port_hdrqfull;
 146         }
 147         if (ipath_stats.sps_etidfull != dd->ipath_last_tidfull) {
 148                 blen += snprintf(buf + blen, sizeof buf - blen,
 149                                  "%srcvegrfull %llu",
 150                                  blen ? ", " : "",
 151                                  (unsigned long long)
 152                                  (ipath_stats.sps_etidfull -
 153                                   dd->ipath_last_tidfull));
 154                 dd->ipath_last_tidfull = ipath_stats.sps_etidfull;
 155         }
 156
 157         /*
 158          * this is actually the number of hdrq full interrupts, not actual
 159          * events, but at the moment that's mostly what I'm interested in.
 160          * Actual count, etc. is in the counters, if needed.  For production
 161          * users this won't ordinarily be printed.
 162          */
 163
 164         if ((ipath_debug & (__IPATH_PKTDBG | __IPATH_DBG)) &&
 165             ipath_stats.sps_hdrqfull != last_tot_hdrqfull) {
 166                 blen += snprintf(buf + blen, sizeof buf - blen,
 167                                  "%shdrqfull %llu (all ports)",
 168                                  blen ? ", " : "",
 169                                  (unsigned long long)
 170                                  (ipath_stats.sps_hdrqfull -
 171                                   last_tot_hdrqfull));
 172                 last_tot_hdrqfull = ipath_stats.sps_hdrqfull;
 173         }
 174         if (blen)
 175                 ipath_dbg("%s\n", buf);
 176
 177         if (dd->ipath_port0head != (u32)
 178             le64_to_cpu(*dd->ipath_hdrqtailptr)) {
 179                 if (dd->ipath_lastport0rcv_cnt ==
 180                     ipath_stats.sps_port0pkts) {
 181                         ipath_cdbg(PKT, "missing rcv interrupts? "
 182                                    "port0 hd=%llx tl=%x; port0pkts %llx\n",
 183                                    (unsigned long long)
 184                                    le64_to_cpu(*dd->ipath_hdrqtailptr),
 185                                    dd->ipath_port0head,
 186                                    (unsigned long long)
 187                                    ipath_stats.sps_port0pkts);
 188                         ipath_kreceive(dd);
 189                 }
 190                 dd->ipath_lastport0rcv_cnt = ipath_stats.sps_port0pkts;
 191         }
 192 }
 193
 194 /**
 195  * ipath_get_faststats - get word counters from chip before they overflow
 196  * @opaque - contains a pointer to the infinipath device ipath_devdata
 197  *
 198  * called from add_timer
 199  */
 200 void ipath_get_faststats(unsigned long opaque)
 201 {
 202         struct ipath_devdata *dd = (struct ipath_devdata *) opaque;
 203         u32 val;
 204         static unsigned cnt;
 205
 206         /*
 207          * don't access the chip while running diags, or memory diags can
 208          * fail
 209          */
 210         if (!dd->ipath_kregbase || !(dd->ipath_flags & IPATH_PRESENT) ||
 211             ipath_diag_inuse)
 212                 /* but re-arm the timer, for diags case; won't hurt other */
 213                 goto done;
 214
 215         if (dd->ipath_flags & IPATH_32BITCOUNTERS) {
 216                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordsendcnt);
 217                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_wordrcvcnt);
 218                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktsendcnt);
 219                 ipath_snap_cntr(dd, dd->ipath_cregs->cr_pktrcvcnt);
 220         }
 221
 222         ipath_qcheck(dd);
 223
 224         /*
 225          * deal with repeat error suppression.  Doesn't really matter if
 226          * last error was almost a full interval ago, or just a few usecs
 227          * ago; still won't get more than 2 per interval.  We may want
 228          * longer intervals for this eventually, could do with mod, counter
 229          * or separate timer.  Also see code in ipath_handle_errors() and
 230          * ipath_handle_hwerrors().
 231          */
 232
 233         if (dd->ipath_lasterror)
 234                 dd->ipath_lasterror = 0;
 235         if (dd->ipath_lasthwerror)
 236                 dd->ipath_lasthwerror = 0;
 237         if ((dd->ipath_maskederrs & ~dd->ipath_ignorederrs)
 238             && time_after(jiffies, dd->ipath_unmasktime)) {
 239                 char ebuf[256];
 240                 ipath_decode_err(ebuf, sizeof ebuf,
 241                                  (dd->ipath_maskederrs & ~dd->
 242                                   ipath_ignorederrs));
 243                 if ((dd->ipath_maskederrs & ~dd->ipath_ignorederrs) &
 244                     ~(INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL))
 245                         ipath_dev_err(dd, "Re-enabling masked errors "
 246                                       "(%s)\n", ebuf);
 247                 else {
 248                         /*
 249                          * rcvegrfull and rcvhdrqfull are "normal", for some
 250                          * types of processes (mostly benchmarks) that send
 251                          * huge numbers of messages, while not processing
 252                          * them.  So only complain about these at debug
 253                          * level.
 254                          */
 255                         ipath_dbg("Disabling frequent queue full errors "
 256                                   "(%s)\n", ebuf);
 257                 }
 258                 dd->ipath_maskederrs = dd->ipath_ignorederrs;
 259                 ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask,
 260                                  ~dd->ipath_maskederrs);
 261         }
 262
 263         /* limit qfull messages to ~one per minute per port */
 264         if ((++cnt & 0x10)) {
 265                 for (val = dd->ipath_cfgports - 1; ((int)val) >= 0;
 266                      val--) {
 267                         if (dd->ipath_lastegrheads[val] != -1)
 268                                 dd->ipath_lastegrheads[val] = -1;
 269                         if (dd->ipath_lastrcvhdrqtails[val] != -1)
 270                                 dd->ipath_lastrcvhdrqtails[val] = -1;
 271                 }
 272         }
 273
 274         if (dd->ipath_nosma_bufs) {
 275                 dd->ipath_nosma_secs += 5;
 276                 if (dd->ipath_nosma_secs >= 30) {
 277                         ipath_cdbg(SMA, "No SMA bufs avail %u seconds; "
 278                                    "cancelling pending sends\n",
 279                                    dd->ipath_nosma_secs);
 280                         /*
 281                          * issue an abort as well, in case we have a packet
 282                          * stuck in launch fifo.  This could corrupt an
 283                          * outgoing user packet in the worst case,
 284                          * but this is a pretty catastrophic, anyway.
 285                          */
 286                         ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
 287                                          INFINIPATH_S_ABORT);
 288                         ipath_disarm_piobufs(dd, dd->ipath_lastport_piobuf,
 289                                              dd->ipath_piobcnt2k +
 290                                              dd->ipath_piobcnt4k -
 291                                              dd->ipath_lastport_piobuf);
 292                         /* start again, if necessary */
 293                         dd->ipath_nosma_secs = 0;
 294                 } else
 295                         ipath_cdbg(SMA, "No SMA bufs avail %u tries, "
 296                                    "after %u seconds\n",
 297                                    dd->ipath_nosma_bufs,
 298                                    dd->ipath_nosma_secs);
 299         }
 300
 301 done:
 302         mod_timer(&dd->ipath_stats_timer, jiffies + HZ * 5);
 303 }