ipath_intr.c 36.9 KB
Newer Older
1
/*
J
John Gregor 已提交
2
 * Copyright (c) 2006, 2007 QLogic Corporation. All rights reserved.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36
 * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
 * General Public License (GPL) Version 2, available from the file
 * COPYING in the main directory of this source tree, or the
 * OpenIB.org BSD license below:
 *
 *     Redistribution and use in source and binary forms, with or
 *     without modification, are permitted provided that the following
 *     conditions are met:
 *
 *      - Redistributions of source code must retain the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer.
 *
 *      - Redistributions in binary form must reproduce the above
 *        copyright notice, this list of conditions and the following
 *        disclaimer in the documentation and/or other materials
 *        provided with the distribution.
 *
 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 * SOFTWARE.
 */

#include <linux/pci.h>

#include "ipath_kernel.h"
37
#include "ipath_verbs.h"
38
#include "ipath_common.h"
39

40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61
/*
 * clear (write) a pio buffer, to clear a parity error.   This routine
 * should only be called when in freeze mode, and the buffer should be
 * canceled afterwards.
 */
static void ipath_clrpiobuf(struct ipath_devdata *dd, u32 pnum)
{
	u32 __iomem *pbuf;
	u32 dwcnt; /* dword count to write */
	if (pnum < dd->ipath_piobcnt2k) {
		pbuf = (u32 __iomem *) (dd->ipath_pio2kbase + pnum *
			dd->ipath_palign);
		dwcnt = dd->ipath_piosize2k >> 2;
	}
	else {
		pbuf = (u32 __iomem *) (dd->ipath_pio4kbase +
			(pnum - dd->ipath_piobcnt2k) * dd->ipath_4kalign);
		dwcnt = dd->ipath_piosize4k >> 2;
	}
	dev_info(&dd->pcidev->dev,
		"Rewrite PIO buffer %u, to recover from parity error\n",
		pnum);
62 63 64 65 66

	/* no flush required, since already in freeze */
	writel(dwcnt + 1, pbuf);
	while (--dwcnt)
		writel(0, pbuf++);
67 68
}

69 70 71
/*
 * Called when we might have an error that is specific to a particular
 * PIO buffer, and may need to cancel that buffer, so it can be re-used.
72 73
 * If rewrite is true, and bits are set in the sendbufferror registers,
 * we'll write to the buffer, for error recovery on parity errors.
74
 */
75
static void ipath_disarm_senderrbufs(struct ipath_devdata *dd, int rewrite)
76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97
{
	u32 piobcnt;
	unsigned long sbuf[4];
	/*
	 * it's possible that sendbuffererror could have bits set; might
	 * have already done this as a result of hardware error handling
	 */
	piobcnt = dd->ipath_piobcnt2k + dd->ipath_piobcnt4k;
	/* read these before writing errorclear */
	sbuf[0] = ipath_read_kreg64(
		dd, dd->ipath_kregs->kr_sendbuffererror);
	sbuf[1] = ipath_read_kreg64(
		dd, dd->ipath_kregs->kr_sendbuffererror + 1);
	if (piobcnt > 128) {
		sbuf[2] = ipath_read_kreg64(
			dd, dd->ipath_kregs->kr_sendbuffererror + 2);
		sbuf[3] = ipath_read_kreg64(
			dd, dd->ipath_kregs->kr_sendbuffererror + 3);
	}

	if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) {
		int i;
98 99
		if (ipath_debug & (__IPATH_PKTDBG|__IPATH_DBG) &&
			dd->ipath_lastcancel > jiffies) {
100 101 102 103 104 105 106 107 108
			__IPATH_DBG_WHICH(__IPATH_PKTDBG|__IPATH_DBG,
					  "SendbufErrs %lx %lx", sbuf[0],
					  sbuf[1]);
			if (ipath_debug & __IPATH_PKTDBG && piobcnt > 128)
				printk(" %lx %lx ", sbuf[2], sbuf[3]);
			printk("\n");
		}

		for (i = 0; i < piobcnt; i++)
109 110 111
			if (test_bit(i, sbuf)) {
				if (rewrite)
					ipath_clrpiobuf(dd, i);
112
				ipath_disarm_piobufs(dd, i, 1);
113
			}
114 115
		/* ignore armlaunch errs for a bit */
		dd->ipath_lastcancel = jiffies+3;
116 117 118 119
	}
}


120
/* These are all rcv-related errors which we want to count for stats */
121 122 123 124 125 126 127 128
#define E_SUM_PKTERRS \
	(INFINIPATH_E_RHDRLEN | INFINIPATH_E_RBADTID | \
	 INFINIPATH_E_RBADVERSION | INFINIPATH_E_RHDR | \
	 INFINIPATH_E_RLONGPKTLEN | INFINIPATH_E_RSHORTPKTLEN | \
	 INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RMINPKTLEN | \
	 INFINIPATH_E_RFORMATERR | INFINIPATH_E_RUNSUPVL | \
	 INFINIPATH_E_RUNEXPCHAR | INFINIPATH_E_REBP)

129
/* These are all send-related errors which we want to count for stats */
130 131 132 133 134 135 136
#define E_SUM_ERRS \
	(INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM | \
	 INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
	 INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SUNSUPVL | \
	 INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SPKTLEN | \
	 INFINIPATH_E_INVALIDADDR)

137 138 139 140 141 142 143 144 145 146 147
/*
 * this is similar to E_SUM_ERRS, but can't ignore armlaunch, don't ignore
 * errors not related to freeze and cancelling buffers.  Can't ignore
 * armlaunch because could get more while still cleaning up, and need
 * to cancel those as they happen.
 */
#define E_SPKT_ERRS_IGNORE \
	 (INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
	 INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SMINPKTLEN | \
	 INFINIPATH_E_SPKTLEN)

148 149 150 151 152 153 154 155 156 157 158 159
/*
 * these are errors that can occur when the link changes state while
 * a packet is being sent or received.  This doesn't cover things
 * like EBP or VCRC that can be the result of a sending having the
 * link change state, so we receive a "known bad" packet.
 */
#define E_SUM_LINK_PKTERRS \
	(INFINIPATH_E_SDROPPEDDATAPKT | INFINIPATH_E_SDROPPEDSMPPKT | \
	 INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SPKTLEN | \
	 INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RMINPKTLEN | \
	 INFINIPATH_E_RUNEXPCHAR)

160 161 162 163
static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs)
{
	u64 ignore_this_time = 0;

164
	ipath_disarm_senderrbufs(dd, 0);
165
	if ((errs & E_SUM_LINK_PKTERRS) &&
166 167 168 169 170 171 172 173
	    !(dd->ipath_flags & IPATH_LINKACTIVE)) {
		/*
		 * This can happen when SMA is trying to bring the link
		 * up, but the IB link changes state at the "wrong" time.
		 * The IB logic then complains that the packet isn't
		 * valid.  We don't want to confuse people, so we just
		 * don't print them, except at debug
		 */
174 175 176
		ipath_dbg("Ignoring packet errors %llx, because link not "
			  "ACTIVE\n", (unsigned long long) errs);
		ignore_this_time = errs & E_SUM_LINK_PKTERRS;
177 178 179 180 181
	}

	return ignore_this_time;
}

182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257
/* generic hw error messages... */
#define INFINIPATH_HWE_TXEMEMPARITYERR_MSG(a) \
	{ \
		.mask = ( INFINIPATH_HWE_TXEMEMPARITYERR_##a <<    \
			  INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT ),   \
		.msg = "TXE " #a " Memory Parity"	     \
	}
#define INFINIPATH_HWE_RXEMEMPARITYERR_MSG(a) \
	{ \
		.mask = ( INFINIPATH_HWE_RXEMEMPARITYERR_##a <<    \
			  INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT ),   \
		.msg = "RXE " #a " Memory Parity"	     \
	}

static const struct ipath_hwerror_msgs ipath_generic_hwerror_msgs[] = {
	INFINIPATH_HWE_MSG(IBCBUSFRSPCPARITYERR, "IPATH2IB Parity"),
	INFINIPATH_HWE_MSG(IBCBUSTOSPCPARITYERR, "IB2IPATH Parity"),

	INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOBUF),
	INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOPBC),
	INFINIPATH_HWE_TXEMEMPARITYERR_MSG(PIOLAUNCHFIFO),

	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(RCVBUF),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(LOOKUPQ),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(EAGERTID),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(EXPTID),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(FLAGBUF),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(DATAINFO),
	INFINIPATH_HWE_RXEMEMPARITYERR_MSG(HDRINFO),
};

/**
 * ipath_format_hwmsg - format a single hwerror message
 * @msg message buffer
 * @msgl length of message buffer
 * @hwmsg message to add to message buffer
 */
static void ipath_format_hwmsg(char *msg, size_t msgl, const char *hwmsg)
{
	strlcat(msg, "[", msgl);
	strlcat(msg, hwmsg, msgl);
	strlcat(msg, "]", msgl);
}

/**
 * ipath_format_hwerrors - format hardware error messages for display
 * @hwerrs hardware errors bit vector
 * @hwerrmsgs hardware error descriptions
 * @nhwerrmsgs number of hwerrmsgs
 * @msg message buffer
 * @msgl message buffer length
 */
void ipath_format_hwerrors(u64 hwerrs,
			   const struct ipath_hwerror_msgs *hwerrmsgs,
			   size_t nhwerrmsgs,
			   char *msg, size_t msgl)
{
	int i;
	const int glen =
	    sizeof(ipath_generic_hwerror_msgs) /
	    sizeof(ipath_generic_hwerror_msgs[0]);

	for (i=0; i<glen; i++) {
		if (hwerrs & ipath_generic_hwerror_msgs[i].mask) {
			ipath_format_hwmsg(msg, msgl,
					   ipath_generic_hwerror_msgs[i].msg);
		}
	}

	for (i=0; i<nhwerrmsgs; i++) {
		if (hwerrs & hwerrmsgs[i].mask) {
			ipath_format_hwmsg(msg, msgl, hwerrmsgs[i].msg);
		}
	}
}

258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279
/* return the strings for the most common link states */
static char *ib_linkstate(u32 linkstate)
{
	char *ret;

	switch (linkstate) {
	case IPATH_IBSTATE_INIT:
		ret = "Init";
		break;
	case IPATH_IBSTATE_ARM:
		ret = "Arm";
		break;
	case IPATH_IBSTATE_ACTIVE:
		ret = "Active";
		break;
	default:
		ret = "Down";
	}

	return ret;
}

280 281 282 283 284 285 286 287 288 289
void signal_ib_event(struct ipath_devdata *dd, enum ib_event_type ev)
{
	struct ib_event event;

	event.device = &dd->verbs_dev->ibdev;
	event.element.port_num = 1;
	event.event = ev;
	ib_dispatch_event(&event);
}

290 291 292 293 294 295 296 297 298 299 300 301 302
static void handle_e_ibstatuschanged(struct ipath_devdata *dd,
				     ipath_err_t errs, int noprint)
{
	u64 val;
	u32 ltstate, lstate;

	/*
	 * even if diags are enabled, we want to notice LINKINIT, etc.
	 * We just don't want to change the LED state, or
	 * dd->ipath_kregs->kr_ibcctrl
	 */
	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
	lstate = val & IPATH_IBSTATE_MASK;
303 304 305 306 307 308 309 310 311 312 313 314 315 316 317

	/*
	 * this is confusing enough when it happens that I want to always put it
	 * on the console and in the logs.  If it was a requested state change,
	 * we'll have already cleared the flags, so we won't print this warning
	 */
	if ((lstate != IPATH_IBSTATE_ARM && lstate != IPATH_IBSTATE_ACTIVE)
		&& (dd->ipath_flags & (IPATH_LINKARMED | IPATH_LINKACTIVE))) {
		dev_info(&dd->pcidev->dev, "Link state changed from %s to %s\n",
				 (dd->ipath_flags & IPATH_LINKARMED) ? "ARM" : "ACTIVE",
				 ib_linkstate(lstate));
		/*
		 * Flush all queued sends when link went to DOWN or INIT,
		 * to be sure that they don't block SMA and other MAD packets
		 */
318
		ipath_cancel_sends(dd, 1);
319 320
	}
	else if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM ||
321 322 323 324 325 326 327 328 329 330 331
	    lstate == IPATH_IBSTATE_ACTIVE) {
		/*
		 * only print at SMA if there is a change, debug if not
		 * (sometimes we want to know that, usually not).
		 */
		if (lstate == ((unsigned) dd->ipath_lastibcstat
			       & IPATH_IBSTATE_MASK)) {
			ipath_dbg("Status change intr but no change (%s)\n",
				  ib_linkstate(lstate));
		}
		else
332
			ipath_cdbg(VERBOSE, "Unit %u link state %s, last "
333 334 335
				   "was %s\n", dd->ipath_unit,
				   ib_linkstate(lstate),
				   ib_linkstate((unsigned)
R
Roland Dreier 已提交
336 337
						dd->ipath_lastibcstat
						& IPATH_IBSTATE_MASK));
338 339 340 341 342 343
	}
	else {
		lstate = dd->ipath_lastibcstat & IPATH_IBSTATE_MASK;
		if (lstate == IPATH_IBSTATE_INIT ||
		    lstate == IPATH_IBSTATE_ARM ||
		    lstate == IPATH_IBSTATE_ACTIVE)
344
			ipath_cdbg(VERBOSE, "Unit %u link state down"
345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387
				   " (state 0x%x), from %s\n",
				   dd->ipath_unit,
				   (u32)val & IPATH_IBSTATE_MASK,
				   ib_linkstate(lstate));
		else
			ipath_cdbg(VERBOSE, "Unit %u link state changed "
				   "to 0x%x from down (%x)\n",
				   dd->ipath_unit, (u32) val, lstate);
	}
	ltstate = (val >> INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) &
		INFINIPATH_IBCS_LINKTRAININGSTATE_MASK;
	lstate = (val >> INFINIPATH_IBCS_LINKSTATE_SHIFT) &
		INFINIPATH_IBCS_LINKSTATE_MASK;

	if (ltstate == INFINIPATH_IBCS_LT_STATE_POLLACTIVE ||
	    ltstate == INFINIPATH_IBCS_LT_STATE_POLLQUIET) {
		u32 last_ltstate;

		/*
		 * Ignore cycling back and forth from Polling.Active
		 * to Polling.Quiet while waiting for the other end of
		 * the link to come up. We will cycle back and forth
		 * between them if no cable is plugged in,
		 * the other device is powered off or disabled, etc.
		 */
		last_ltstate = (dd->ipath_lastibcstat >>
				INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT)
			& INFINIPATH_IBCS_LINKTRAININGSTATE_MASK;
		if (last_ltstate == INFINIPATH_IBCS_LT_STATE_POLLACTIVE
		    || last_ltstate ==
		    INFINIPATH_IBCS_LT_STATE_POLLQUIET) {
			if (dd->ipath_ibpollcnt > 40) {
				dd->ipath_flags |= IPATH_NOCABLE;
				*dd->ipath_statusp |=
					IPATH_STATUS_IB_NOCABLE;
			} else
				dd->ipath_ibpollcnt++;
			goto skip_ibchange;
		}
	}
	dd->ipath_ibpollcnt = 0;	/* some state other than 2 or 3 */
	ipath_stats.sps_iblink++;
	if (ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) {
388 389
		if (dd->ipath_flags & IPATH_LINKACTIVE)
			signal_ib_event(dd, IB_EVENT_PORT_ERR);
390 391 392 393 394
		dd->ipath_flags |= IPATH_LINKDOWN;
		dd->ipath_flags &= ~(IPATH_LINKUNK | IPATH_LINKINIT
				     | IPATH_LINKACTIVE |
				     IPATH_LINKARMED);
		*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
395
		dd->ipath_lli_counter = 0;
396 397 398 399 400 401
		if (!noprint) {
			if (((dd->ipath_lastibcstat >>
			      INFINIPATH_IBCS_LINKSTATE_SHIFT) &
			     INFINIPATH_IBCS_LINKSTATE_MASK)
			    == INFINIPATH_IBCS_L_STATE_ACTIVE)
				/* if from up to down be more vocal */
402
				ipath_cdbg(VERBOSE,
403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421
					   "Unit %u link now down (%s)\n",
					   dd->ipath_unit,
					   ipath_ibcstatus_str[ltstate]);
			else
				ipath_cdbg(VERBOSE, "Unit %u link is "
					   "down (%s)\n", dd->ipath_unit,
					   ipath_ibcstatus_str[ltstate]);
		}

		dd->ipath_f_setextled(dd, lstate, ltstate);
	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_ACTIVE) {
		dd->ipath_flags |= IPATH_LINKACTIVE;
		dd->ipath_flags &=
			~(IPATH_LINKUNK | IPATH_LINKINIT | IPATH_LINKDOWN |
			  IPATH_LINKARMED | IPATH_NOCABLE);
		*dd->ipath_statusp &= ~IPATH_STATUS_IB_NOCABLE;
		*dd->ipath_statusp |=
			IPATH_STATUS_IB_READY | IPATH_STATUS_IB_CONF;
		dd->ipath_f_setextled(dd, lstate, ltstate);
422
		signal_ib_event(dd, IB_EVENT_PORT_ACTIVE);
423
	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) {
424 425
		if (dd->ipath_flags & IPATH_LINKACTIVE)
			signal_ib_event(dd, IB_EVENT_PORT_ERR);
426 427 428 429 430 431 432 433 434 435 436 437
		/*
		 * set INIT and DOWN.  Down is checked by most of the other
		 * code, but INIT is useful to know in a few places.
		 */
		dd->ipath_flags |= IPATH_LINKINIT | IPATH_LINKDOWN;
		dd->ipath_flags &=
			~(IPATH_LINKUNK | IPATH_LINKACTIVE | IPATH_LINKARMED
			  | IPATH_NOCABLE);
		*dd->ipath_statusp &= ~(IPATH_STATUS_IB_NOCABLE
					| IPATH_STATUS_IB_READY);
		dd->ipath_f_setextled(dd, lstate, ltstate);
	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_ARM) {
438 439
		if (dd->ipath_flags & IPATH_LINKACTIVE)
			signal_ib_event(dd, IB_EVENT_PORT_ERR);
440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457
		dd->ipath_flags |= IPATH_LINKARMED;
		dd->ipath_flags &=
			~(IPATH_LINKUNK | IPATH_LINKDOWN | IPATH_LINKINIT |
			  IPATH_LINKACTIVE | IPATH_NOCABLE);
		*dd->ipath_statusp &= ~(IPATH_STATUS_IB_NOCABLE
					| IPATH_STATUS_IB_READY);
		dd->ipath_f_setextled(dd, lstate, ltstate);
	} else {
		if (!noprint)
			ipath_dbg("IBstatuschange unit %u: %s (%x)\n",
				  dd->ipath_unit,
				  ipath_ibcstatus_str[ltstate], ltstate);
	}
skip_ibchange:
	dd->ipath_lastibcstat = val;
}

static void handle_supp_msgs(struct ipath_devdata *dd,
458
			     unsigned supp_msgs, char *msg, int msgsz)
459 460 461 462 463 464
{
	/*
	 * Print the message unless it's ibc status change only, which
	 * happens so often we never want to count it.
	 */
	if (dd->ipath_lasterror & ~INFINIPATH_E_IBSTATUSCHANGED) {
465
		int iserr;
466 467 468
		iserr = ipath_decode_err(msg, msgsz,
					 dd->ipath_lasterror &
					 ~INFINIPATH_E_IBSTATUSCHANGED);
469
		if (dd->ipath_lasterror &
470 471
			~(INFINIPATH_E_RRCVEGRFULL |
			INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS))
472 473 474 475 476 477 478 479 480 481 482 483 484
			ipath_dev_err(dd, "Suppressed %u messages for "
				      "fast-repeating errors (%s) (%llx)\n",
				      supp_msgs, msg,
				      (unsigned long long)
				      dd->ipath_lasterror);
		else {
			/*
			 * rcvegrfull and rcvhdrqfull are "normal", for some
			 * types of processes (mostly benchmarks) that send
			 * huge numbers of messages, while not processing
			 * them. So only complain about these at debug
			 * level.
			 */
485 486 487 488 489 490 491
			if (iserr)
				ipath_dbg("Suppressed %u messages for %s\n",
					  supp_msgs, msg);
			else
				ipath_cdbg(ERRPKT,
					"Suppressed %u messages for %s\n",
					  supp_msgs, msg);
492 493 494 495 496
		}
	}
}

static unsigned handle_frequent_errors(struct ipath_devdata *dd,
497 498
				       ipath_err_t errs, char *msg,
				       int msgsz, int *noprint)
499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516
{
	unsigned long nc;
	static unsigned long nextmsg_time;
	static unsigned nmsgs, supp_msgs;

	/*
	 * Throttle back "fast" messages to no more than 10 per 5 seconds.
	 * This isn't perfect, but it's a reasonable heuristic. If we get
	 * more than 10, give a 6x longer delay.
	 */
	nc = jiffies;
	if (nmsgs > 10) {
		if (time_before(nc, nextmsg_time)) {
			*noprint = 1;
			if (!supp_msgs++)
				nextmsg_time = nc + HZ * 3;
		}
		else if (supp_msgs) {
517
			handle_supp_msgs(dd, supp_msgs, msg, msgsz);
518 519 520 521 522 523 524 525 526 527
			supp_msgs = 0;
			nmsgs = 0;
		}
	}
	else if (!nmsgs++ || time_after(nc, nextmsg_time))
		nextmsg_time = nc + HZ / 2;

	return supp_msgs;
}

528
static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
529
{
530
	char msg[128];
531
	u64 ignore_this_time = 0;
532
	int i, iserr = 0;
533 534
	int chkerrpkts = 0, noprint = 0;
	unsigned supp_msgs;
535
	int log_idx;
536

537
	supp_msgs = handle_frequent_errors(dd, errs, msg, sizeof msg, &noprint);
538

539
	/* don't report errors that are masked */
540 541 542 543 544 545
	errs &= ~dd->ipath_maskederrs;

	/* do these first, they are most important */
	if (errs & INFINIPATH_E_HARDWARE) {
		/* reuse same msg buf */
		dd->ipath_f_handle_hwerrors(dd, msg, sizeof msg);
546 547 548 549 550 551 552
	} else {
		u64 mask;
		for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx) {
			mask = dd->ipath_eep_st_masks[log_idx].errs_to_log;
			if (errs & mask)
				ipath_inc_eeprom_err(dd, log_idx, 1);
		}
553 554
	}

555
	if (!noprint && (errs & ~dd->ipath_e_bitsextant))
556 557
		ipath_dev_err(dd, "error interrupt with unknown errors "
			      "%llx set\n", (unsigned long long)
558
			      (errs & ~dd->ipath_e_bitsextant));
559 560 561

	if (errs & E_SUM_ERRS)
		ignore_this_time = handle_e_sum_errs(dd, errs);
562 563 564 565 566 567 568 569 570 571 572 573 574
	else if ((errs & E_SUM_LINK_PKTERRS) &&
	    !(dd->ipath_flags & IPATH_LINKACTIVE)) {
		/*
		 * This can happen when SMA is trying to bring the link
		 * up, but the IB link changes state at the "wrong" time.
		 * The IB logic then complains that the packet isn't
		 * valid.  We don't want to confuse people, so we just
		 * don't print them, except at debug
		 */
		ipath_dbg("Ignoring packet errors %llx, because link not "
			  "ACTIVE\n", (unsigned long long) errs);
		ignore_this_time = errs & E_SUM_LINK_PKTERRS;
	}
575 576

	if (supp_msgs == 250000) {
577
		int s_iserr;
578 579 580 581 582 583 584
		/*
		 * It's not entirely reasonable assuming that the errors set
		 * in the last clear period are all responsible for the
		 * problem, but the alternative is to assume it's the only
		 * ones on this particular interrupt, which also isn't great
		 */
		dd->ipath_maskederrs |= dd->ipath_lasterror | errs;
585
		dd->ipath_errormask &= ~dd->ipath_maskederrs;
586
		ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask,
587
			dd->ipath_errormask);
588
		s_iserr = ipath_decode_err(msg, sizeof msg,
589
			dd->ipath_maskederrs);
590

591
		if (dd->ipath_maskederrs &
592 593 594 595
			~(INFINIPATH_E_RRCVEGRFULL |
			INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS))
			ipath_dev_err(dd, "Temporarily disabling "
			    "error(s) %llx reporting; too frequent (%s)\n",
596 597
				(unsigned long long)dd->ipath_maskederrs,
				msg);
598 599 600 601 602 603 604 605
		else {
			/*
			 * rcvegrfull and rcvhdrqfull are "normal",
			 * for some types of processes (mostly benchmarks)
			 * that send huge numbers of messages, while not
			 * processing them.  So only complain about
			 * these at debug level.
			 */
606 607 608 609 610 611 612 613 614
			if (s_iserr)
				ipath_dbg("Temporarily disabling reporting "
				    "too frequent queue full errors (%s)\n",
				    msg);
			else
				ipath_cdbg(ERRPKT,
				    "Temporarily disabling reporting too"
				    " frequent packet errors (%s)\n",
				    msg);
615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636
		}

		/*
		 * Re-enable the masked errors after around 3 minutes.  in
		 * ipath_get_faststats().  If we have a series of fast
		 * repeating but different errors, the interval will keep
		 * stretching out, but that's OK, as that's pretty
		 * catastrophic.
		 */
		dd->ipath_unmasktime = jiffies + HZ * 180;
	}

	ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear, errs);
	if (ignore_this_time)
		errs &= ~ignore_this_time;
	if (errs & ~dd->ipath_lasterror) {
		errs &= ~dd->ipath_lasterror;
		/* never suppress duplicate hwerrors or ibstatuschange */
		dd->ipath_lasterror |= errs &
			~(INFINIPATH_E_HARDWARE |
			  INFINIPATH_E_IBSTATUSCHANGED);
	}
637 638 639 640 641 642 643 644

	/* likely due to cancel, so suppress */
	if ((errs & (INFINIPATH_E_SPKTLEN | INFINIPATH_E_SPIOARMLAUNCH)) &&
		dd->ipath_lastcancel > jiffies) {
		ipath_dbg("Suppressed armlaunch/spktlen after error send cancel\n");
		errs &= ~(INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SPKTLEN);
	}

645
	if (!errs)
646
		return 0;
647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671

	if (!noprint)
		/*
		 * the ones we mask off are handled specially below or above
		 */
		ipath_decode_err(msg, sizeof msg,
				 errs & ~(INFINIPATH_E_IBSTATUSCHANGED |
					  INFINIPATH_E_RRCVEGRFULL |
					  INFINIPATH_E_RRCVHDRFULL |
					  INFINIPATH_E_HARDWARE));
	else
		/* so we don't need if (!noprint) at strlcat's below */
		*msg = 0;

	if (errs & E_SUM_PKTERRS) {
		ipath_stats.sps_pkterrs++;
		chkerrpkts = 1;
	}
	if (errs & E_SUM_ERRS)
		ipath_stats.sps_errs++;

	if (errs & (INFINIPATH_E_RICRC | INFINIPATH_E_RVCRC)) {
		ipath_stats.sps_crcerrs++;
		chkerrpkts = 1;
	}
672 673
	iserr = errs & ~(E_SUM_PKTERRS | INFINIPATH_E_PKTERRS);

674 675 676 677 678 679 680 681 682 683 684

	/*
	 * We don't want to print these two as they happen, or we can make
	 * the situation even worse, because it takes so long to print
	 * messages to serial consoles.  Kernel ports get printed from
	 * fast_stats, no more than every 5 seconds, user ports get printed
	 * on close
	 */
	if (errs & INFINIPATH_E_RRCVHDRFULL) {
		u32 hd, tl;
		ipath_stats.sps_hdrqfull++;
685
		for (i = 0; i < dd->ipath_cfgports; i++) {
686 687
			struct ipath_portdata *pd = dd->ipath_pd[i];
			if (i == 0) {
688
				hd = pd->port_head;
689 690 691 692 693 694 695 696
				tl = (u32) le64_to_cpu(
					*dd->ipath_hdrqtailptr);
			} else if (pd && pd->port_cnt &&
				   pd->port_rcvhdrtail_kvaddr) {
				/*
				 * don't report same point multiple times,
				 * except kernel
				 */
697
				tl = *(u64 *) pd->port_rcvhdrtail_kvaddr;
698
				if (tl == pd->port_lastrcvhdrqtail)
699 700 701 702 703 704 705 706 707
					continue;
				hd = ipath_read_ureg32(dd, ur_rcvhdrhead,
						       i);
			} else
				continue;
			if (hd == (tl + 1) ||
			    (!hd && tl == dd->ipath_hdrqlast)) {
				if (i == 0)
					chkerrpkts = 1;
708
				pd->port_lastrcvhdrqtail = tl;
709
				pd->port_hdrqfull++;
710 711 712
				/* flush hdrqfull so that poll() sees it */
				wmb();
				wake_up_interruptible(&pd->port_wait);
713 714 715 716
			}
		}
	}
	if (errs & INFINIPATH_E_RRCVEGRFULL) {
717 718
		struct ipath_portdata *pd = dd->ipath_pd[0];

719 720 721 722 723 724 725
		/*
		 * since this is of less importance and not likely to
		 * happen without also getting hdrfull, only count
		 * occurrences; don't check each port (or even the kernel
		 * vs user)
		 */
		ipath_stats.sps_etidfull++;
726
		if (pd->port_head !=
727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766
		    (u32) le64_to_cpu(*dd->ipath_hdrqtailptr))
			chkerrpkts = 1;
	}

	/*
	 * do this before IBSTATUSCHANGED, in case both bits set in a single
	 * interrupt; we want the STATUSCHANGE to "win", so we do our
	 * internal copy of state machine correctly
	 */
	if (errs & INFINIPATH_E_RIBLOSTLINK) {
		/*
		 * force through block below
		 */
		errs |= INFINIPATH_E_IBSTATUSCHANGED;
		ipath_stats.sps_iblink++;
		dd->ipath_flags |= IPATH_LINKDOWN;
		dd->ipath_flags &= ~(IPATH_LINKUNK | IPATH_LINKINIT
				     | IPATH_LINKARMED | IPATH_LINKACTIVE);
		*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
		if (!noprint) {
			u64 st = ipath_read_kreg64(
				dd, dd->ipath_kregs->kr_ibcstatus);

			ipath_dbg("Lost link, link now down (%s)\n",
				  ipath_ibcstatus_str[st & 0xf]);
		}
	}
	if (errs & INFINIPATH_E_IBSTATUSCHANGED)
		handle_e_ibstatuschanged(dd, errs, noprint);

	if (errs & INFINIPATH_E_RESET) {
		if (!noprint)
			ipath_dev_err(dd, "Got reset, requires re-init "
				      "(unload and reload driver)\n");
		dd->ipath_flags &= ~IPATH_INITTED;	/* needs re-init */
		/* mark as having had error */
		*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
		*dd->ipath_statusp &= ~IPATH_STATUS_IB_CONF;
	}

767 768 769 770 771 772 773
	if (!noprint && *msg) {
		if (iserr)
			ipath_dev_err(dd, "%s error\n", msg);
		else
			dev_info(&dd->pcidev->dev, "%s packet problems\n",
				msg);
	}
774 775 776
	if (dd->ipath_state_wanted & dd->ipath_flags) {
		ipath_cdbg(VERBOSE, "driver wanted state %x, iflags now %x, "
			   "waking\n", dd->ipath_state_wanted,
777
			   dd->ipath_flags);
778
		wake_up_interruptible(&ipath_state_wait);
779 780
	}

781
	return chkerrpkts;
782 783
}

784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800

/*
 * try to cleanup as much as possible for anything that might have gone
 * wrong while in freeze mode, such as pio buffers being written by user
 * processes (causing armlaunch), send errors due to going into freeze mode,
 * etc., and try to avoid causing extra interrupts while doing so.
 * Forcibly update the in-memory pioavail register copies after cleanup
 * because the chip won't do it for anything changing while in freeze mode
 * (we don't want to wait for the next pio buffer state change).
 * Make sure that we don't lose any important interrupts by using the chip
 * feature that says that writing 0 to a bit in *clear that is set in
 * *status will cause an interrupt to be generated again (if allowed by
 * the *mask value).
 */
void ipath_clear_freeze(struct ipath_devdata *dd)
{
	int i, im;
801
	u64 val;
J
John Gregor 已提交
802
	unsigned long flags;
803 804 805 806

	/* disable error interrupts, to avoid confusion */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask, 0ULL);

807 808 809
	/* also disable interrupts; errormask is sometimes overwriten */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, 0ULL);

810 811 812 813 814 815
	/*
	 * clear all sends, because they have may been
	 * completed by usercode while in freeze mode, and
	 * therefore would not be sent, and eventually
	 * might cause the process to run out of bufs
	 */
816
	ipath_cancel_sends(dd, 0);
817 818 819 820
	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
			 dd->ipath_control);

	/* ensure pio avail updates continue */
J
John Gregor 已提交
821
	spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags);
822
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
823
		 dd->ipath_sendctrl & ~INFINIPATH_S_PIOBUFAVAILUPD);
824 825
	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
J
John Gregor 已提交
826 827 828
			 dd->ipath_sendctrl);
	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags);
829 830 831 832 833 834 835

	/*
	 * We just enabled pioavailupdate, so dma copy is almost certainly
	 * not yet right, so read the registers directly.  Similar to init
	 */
	for (i = 0; i < dd->ipath_pioavregs; i++) {
		/* deal with 6110 chip bug */
836 837
		im = (i > 3 && (dd->ipath_flags & IPATH_SWAP_PIOBUFS)) ?
			i ^ 1 : i;
838
		val = ipath_read_kreg64(dd, (0x1000 / sizeof(u64)) + im);
839 840
		dd->ipath_pioavailregs_dma[i] = cpu_to_le64(val);
		dd->ipath_pioavailshadow[i] = val;
841 842 843 844 845 846 847 848 849 850 851 852
	}

	/*
	 * force new interrupt if any hwerr, error or interrupt bits are
	 * still set, and clear "safe" send packet errors related to freeze
	 * and cancelling sends.  Re-enable error interrupts before possible
	 * force of re-interrupt on pending interrupts.
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear, 0ULL);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_errorclear,
		E_SPKT_ERRS_IGNORE);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask,
853 854
		dd->ipath_errormask);
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, -1LL);
855 856 857 858
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, 0ULL);
}


859 860
/* this is separate to allow for better optimization of ipath_intr() */

861
static noinline void ipath_bad_intr(struct ipath_devdata *dd, u32 *unexpectp)
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878
{
	/*
	 * sometimes happen during driver init and unload, don't want
	 * to process any interrupts at that point
	 */

	/* this is just a bandaid, not a fix, if something goes badly
	 * wrong */
	if (++*unexpectp > 100) {
		if (++*unexpectp > 105) {
			/*
			 * ok, we must be taking somebody else's interrupts,
			 * due to a messed up mptable and/or PIRQ table, so
			 * unregister the interrupt.  We've seen this during
			 * linuxbios development work, and it may happen in
			 * the future again.
			 */
879
			if (dd->pcidev && dd->ipath_irq) {
880 881 882 883
				ipath_dev_err(dd, "Now %u unexpected "
					      "interrupts, unregistering "
					      "interrupt handler\n",
					      *unexpectp);
884 885 886
				ipath_dbg("free_irq of irq %d\n",
					  dd->ipath_irq);
				dd->ipath_f_free_irq(dd);
887 888
			}
		}
889
		if (ipath_read_ireg(dd, dd->ipath_kregs->kr_intmask)) {
890 891 892 893 894 895 896 897 898 899 900 901 902 903
			ipath_dev_err(dd, "%u unexpected interrupts, "
				      "disabling interrupts completely\n",
				      *unexpectp);
			/*
			 * disable all interrupts, something is very wrong
			 */
			ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask,
					 0ULL);
		}
	} else if (*unexpectp > 1)
		ipath_dbg("Interrupt when not ready, should not happen, "
			  "ignoring\n");
}

904
static noinline void ipath_bad_regread(struct ipath_devdata *dd)
905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921
{
	static int allbits;

	/* separate routine, for better optimization of ipath_intr() */

	/*
	 * We print the message and disable interrupts, in hope of
	 * having a better chance of debugging the problem.
	 */
	ipath_dev_err(dd,
		      "Read of interrupt status failed (all bits set)\n");
	if (allbits++) {
		/* disable all interrupts, something is very wrong */
		ipath_write_kreg(dd, dd->ipath_kregs->kr_intmask, 0ULL);
		if (allbits == 2) {
			ipath_dev_err(dd, "Still bad interrupt status, "
				      "unregistering interrupt\n");
922
			dd->ipath_f_free_irq(dd);
923 924 925 926 927 928 929 930 931 932 933
		} else if (allbits > 2) {
			if ((allbits % 10000) == 0)
				printk(".");
		} else
			ipath_dev_err(dd, "Disabling interrupts, "
				      "multiple errors\n");
	}
}

static void handle_layer_pioavail(struct ipath_devdata *dd)
{
J
John Gregor 已提交
934
	unsigned long flags;
935 936
	int ret;

937
	ret = ipath_ib_piobufavail(dd->verbs_dev);
938
	if (ret > 0)
939
		goto set;
940 941

	return;
942
set:
J
John Gregor 已提交
943 944
	spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags);
	dd->ipath_sendctrl |= INFINIPATH_S_PIOINTBUFAVAIL;
945 946
	ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
			 dd->ipath_sendctrl);
J
John Gregor 已提交
947 948
	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
	spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags);
949 950
}

951 952 953 954 955 956
/*
 * Handle receive interrupts for user ports; this means a user
 * process was waiting for a packet to arrive, and didn't want
 * to poll
 */
static void handle_urcv(struct ipath_devdata *dd, u32 istat)
957 958 959 960 961
{
	u64 portr;
	int i;
	int rcvdint = 0;

962 963 964 965 966 967 968 969 970
	/*
	 * test_and_clear_bit(IPATH_PORT_WAITING_RCV) and
	 * test_and_clear_bit(IPATH_PORT_WAITING_URG) below
	 * would both like timely updates of the bits so that
	 * we don't pass them by unnecessarily.  the rmb()
	 * here ensures that we see them promptly -- the
	 * corresponding wmb()'s are in ipath_poll_urgent()
	 * and ipath_poll_next()...
	 */
971
	rmb();
972
	portr = ((istat >> INFINIPATH_I_RCVAVAIL_SHIFT) &
973
		 dd->ipath_i_rcvavail_mask)
974
		| ((istat >> INFINIPATH_I_RCVURG_SHIFT) &
975
		   dd->ipath_i_rcvurg_mask);
976
	for (i = 1; i < dd->ipath_cfgports; i++) {
977
		struct ipath_portdata *pd = dd->ipath_pd[i];
978
		if (portr & (1 << i) && pd && pd->port_cnt) {
979 980
			if (test_and_clear_bit(IPATH_PORT_WAITING_RCV,
					       &pd->port_flag)) {
981
				clear_bit(i + dd->ipath_r_intravail_shift,
982 983 984
					  &dd->ipath_rcvctrl);
				wake_up_interruptible(&pd->port_wait);
				rcvdint = 1;
985 986 987
			} else if (test_and_clear_bit(IPATH_PORT_WAITING_URG,
						      &pd->port_flag)) {
				pd->port_urgent++;
988 989
				wake_up_interruptible(&pd->port_wait);
			}
990 991 992 993 994 995 996 997 998 999 1000 1001
		}
	}
	if (rcvdint) {
		/* only want to take one interrupt, so turn off the rcv
		 * interrupt for all the ports that we did the wakeup on
		 * (but never for kernel port)
		 */
		ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
				 dd->ipath_rcvctrl);
	}
}

1002
irqreturn_t ipath_intr(int irq, void *data)
1003 1004
{
	struct ipath_devdata *dd = data;
1005
	u32 istat, chk0rcv = 0;
1006 1007
	ipath_err_t estat = 0;
	irqreturn_t ret;
1008 1009 1010 1011 1012
	static unsigned unexpected = 0;
	static const u32 port0rbits = (1U<<INFINIPATH_I_RCVAVAIL_SHIFT) |
		 (1U<<INFINIPATH_I_RCVURG_SHIFT);

	ipath_stats.sps_ints++;
1013

1014 1015 1016
	if (dd->ipath_int_counter != (u32) -1)
		dd->ipath_int_counter++;

1017
	if (!(dd->ipath_flags & IPATH_PRESENT)) {
1018
		/*
1019
		 * This return value is not great, but we do not want the
1020 1021 1022 1023 1024 1025 1026
		 * interrupt core code to remove our interrupt handler
		 * because we don't appear to be handling an interrupt
		 * during a chip reset.
		 */
		return IRQ_HANDLED;
	}

1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039
	/*
	 * this needs to be flags&initted, not statusp, so we keep
	 * taking interrupts even after link goes down, etc.
	 * Also, we *must* clear the interrupt at some point, or we won't
	 * take it again, which can be real bad for errors, etc...
	 */

	if (!(dd->ipath_flags & IPATH_INITTED)) {
		ipath_bad_intr(dd, &unexpected);
		ret = IRQ_NONE;
		goto bail;
	}

1040
	istat = ipath_read_ireg(dd, dd->ipath_kregs->kr_intstatus);
1041

1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056
	if (unlikely(!istat)) {
		ipath_stats.sps_nullintr++;
		ret = IRQ_NONE; /* not our interrupt, or already handled */
		goto bail;
	}
	if (unlikely(istat == -1)) {
		ipath_bad_regread(dd);
		/* don't know if it was our interrupt or not */
		ret = IRQ_NONE;
		goto bail;
	}

	if (unexpected)
		unexpected = 0;

1057
	if (unlikely(istat & ~dd->ipath_i_bitsextant))
1058 1059
		ipath_dev_err(dd,
			      "interrupt with unknown interrupts %x set\n",
1060
			      istat & (u32) ~ dd->ipath_i_bitsextant);
1061 1062
	else
		ipath_cdbg(VERBOSE, "intr stat=0x%x\n", istat);
1063

1064
	if (unlikely(istat & INFINIPATH_I_ERROR)) {
1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078
		ipath_stats.sps_errints++;
		estat = ipath_read_kreg64(dd,
					  dd->ipath_kregs->kr_errorstatus);
		if (!estat)
			dev_info(&dd->pcidev->dev, "error interrupt (%x), "
				 "but no error bits set!\n", istat);
		else if (estat == -1LL)
			/*
			 * should we try clearing all, or hope next read
			 * works?
			 */
			ipath_dev_err(dd, "Read of error status failed "
				      "(all bits set); ignoring\n");
		else
1079 1080 1081
			if (handle_errors(dd, estat))
				/* force calling ipath_kreceive() */
				chk0rcv = 1;
1082 1083 1084
	}

	if (istat & INFINIPATH_I_GPIO) {
1085
		/*
1086 1087 1088 1089
		 * GPIO interrupts fall in two broad classes:
		 * GPIO_2 indicates (on some HT4xx boards) that a packet
		 *        has arrived for Port 0. Checking for this
		 *        is controlled by flag IPATH_GPIO_INTR.
1090 1091
		 * GPIO_3..5 on IBA6120 Rev2 and IBA6110 Rev4 chips indicate
		 *        errors that we need to count. Checking for this
1092
		 *        is controlled by flag IPATH_GPIO_ERRINTRS.
1093
		 */
1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122
		u32 gpiostatus;
		u32 to_clear = 0;

		gpiostatus = ipath_read_kreg32(
			dd, dd->ipath_kregs->kr_gpio_status);
		/* First the error-counter case.
		 */
		if ((gpiostatus & IPATH_GPIO_ERRINTR_MASK) &&
		    (dd->ipath_flags & IPATH_GPIO_ERRINTRS)) {
			/* want to clear the bits we see asserted. */
			to_clear |= (gpiostatus & IPATH_GPIO_ERRINTR_MASK);

			/*
			 * Count appropriately, clear bits out of our copy,
			 * as they have been "handled".
			 */
			if (gpiostatus & (1 << IPATH_GPIO_RXUVL_BIT)) {
				ipath_dbg("FlowCtl on UnsupVL\n");
				dd->ipath_rxfc_unsupvl_errs++;
			}
			if (gpiostatus & (1 << IPATH_GPIO_OVRUN_BIT)) {
				ipath_dbg("Overrun Threshold exceeded\n");
				dd->ipath_overrun_thresh_errs++;
			}
			if (gpiostatus & (1 << IPATH_GPIO_LLI_BIT)) {
				ipath_dbg("Local Link Integrity error\n");
				dd->ipath_lli_errs++;
			}
			gpiostatus &= ~IPATH_GPIO_ERRINTR_MASK;
1123
		}
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135
		/* Now the Port0 Receive case */
		if ((gpiostatus & (1 << IPATH_GPIO_PORT0_BIT)) &&
		    (dd->ipath_flags & IPATH_GPIO_INTR)) {
			/*
			 * GPIO status bit 2 is set, and we expected it.
			 * clear it and indicate in p0bits.
			 * This probably only happens if a Port0 pkt
			 * arrives at _just_ the wrong time, and we
			 * handle that by seting chk0rcv;
			 */
			to_clear |= (1 << IPATH_GPIO_PORT0_BIT);
			gpiostatus &= ~(1 << IPATH_GPIO_PORT0_BIT);
1136
			chk0rcv = 1;
1137
		}
1138
		if (gpiostatus) {
1139 1140 1141
			/*
			 * Some unexpected bits remain. If they could have
			 * caused the interrupt, complain and clear.
1142 1143
			 * To avoid repetition of this condition, also clear
			 * the mask. It is almost certainly due to error.
1144
			 */
1145 1146
			const u32 mask = (u32) dd->ipath_gpio_mask;

1147 1148 1149 1150
			if (mask & gpiostatus) {
				ipath_dbg("Unexpected GPIO IRQ bits %x\n",
				  gpiostatus & mask);
				to_clear |= (gpiostatus & mask);
1151 1152 1153 1154
				dd->ipath_gpio_mask &= ~(gpiostatus & mask);
				ipath_write_kreg(dd,
					dd->ipath_kregs->kr_gpio_mask,
					dd->ipath_gpio_mask);
1155 1156 1157 1158 1159 1160
			}
		}
		if (to_clear) {
			ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_clear,
					(u64) to_clear);
		}
1161
	}
1162
	chk0rcv |= istat & port0rbits;
1163 1164

	/*
1165 1166 1167 1168 1169 1170
	 * Clear the interrupt bits we found set, unless they are receive
	 * related, in which case we already cleared them above, and don't
	 * want to clear them again, because we might lose an interrupt.
	 * Clear it early, so we "know" know the chip will have seen this by
	 * the time we process the queue, and will re-interrupt if necessary.
	 * The processor itself won't take the interrupt again until we return.
1171 1172 1173
	 */
	ipath_write_kreg(dd, dd->ipath_kregs->kr_intclear, istat);

1174
	/*
1175 1176 1177 1178
	 * handle port0 receive  before checking for pio buffers available,
	 * since receives can overflow; piobuf waiters can afford a few
	 * extra cycles, since they were waiting anyway, and user's waiting
	 * for receive are at the bottom.
1179
	 */
1180
	if (chk0rcv) {
1181
		ipath_kreceive(dd->ipath_pd[0]);
1182 1183
		istat &= ~port0rbits;
	}
1184

1185
	if (istat & ((dd->ipath_i_rcvavail_mask <<
1186
		      INFINIPATH_I_RCVAVAIL_SHIFT)
1187
		     | (dd->ipath_i_rcvurg_mask <<
1188 1189 1190
			INFINIPATH_I_RCVURG_SHIFT)))
		handle_urcv(dd, istat);

1191
	if (istat & INFINIPATH_I_SPIOBUFAVAIL) {
J
John Gregor 已提交
1192 1193 1194 1195
		unsigned long flags;

		spin_lock_irqsave(&dd->ipath_sendctrl_lock, flags);
		dd->ipath_sendctrl &= ~INFINIPATH_S_PIOINTBUFAVAIL;
1196 1197
		ipath_write_kreg(dd, dd->ipath_kregs->kr_sendctrl,
				 dd->ipath_sendctrl);
J
John Gregor 已提交
1198 1199
		ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
		spin_unlock_irqrestore(&dd->ipath_sendctrl_lock, flags);
1200 1201 1202 1203 1204 1205 1206 1207 1208

		handle_layer_pioavail(dd);
	}

	ret = IRQ_HANDLED;

bail:
	return ret;
}