ipath_intr.c

来自「LINUX 2.6.17.4的源码」· C语言 代码 · 共 855 行 · 第 1/2 页

/*
 * 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"
#include "ips_common.h"
#include "ipath_layer.h"

#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)

#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)

static u64 handle_e_sum_errs(struct ipath_devdata *dd, ipath_err_t errs)
{
	unsigned long sbuf[4];
	u64 ignore_this_time = 0;
	u32 piobcnt;

	/* if possible that sendbuffererror could be valid */
	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;

		ipath_cdbg(PKT, "SendbufErrs %lx %lx ", sbuf[0], sbuf[1]);
		if (ipath_debug & __IPATH_PKTDBG && piobcnt > 128)
			printk("%lx %lx ", sbuf[2], sbuf[3]);
		for (i = 0; i < piobcnt; i++) {
			if (test_bit(i, sbuf)) {
				u32 __iomem *piobuf;
				if (i < dd->ipath_piobcnt2k)
					piobuf = (u32 __iomem *)
						(dd->ipath_pio2kbase +
						 i * dd->ipath_palign);
				else
					piobuf = (u32 __iomem *)
						(dd->ipath_pio4kbase +
						 (i - dd->ipath_piobcnt2k) *
						 dd->ipath_4kalign);

				ipath_cdbg(PKT,
					   "PIObuf[%u] @%p pbc is %x; ",
					   i, piobuf, readl(piobuf));

				ipath_disarm_piobufs(dd, i, 1);
			}
		}
		if (ipath_debug & __IPATH_PKTDBG)
			printk("\n");
	}
	if ((errs & (INFINIPATH_E_SDROPPEDDATAPKT |
		     INFINIPATH_E_SDROPPEDSMPPKT |
		     INFINIPATH_E_SMINPKTLEN)) &&
	    !(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 pktsend errors %llx, because not "
			  "yet active\n", (unsigned long long) errs);
		ignore_this_time = INFINIPATH_E_SDROPPEDDATAPKT |
			INFINIPATH_E_SDROPPEDSMPPKT |
			INFINIPATH_E_SMINPKTLEN;
	}

	return ignore_this_time;
}

/* 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;
}

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;
	if (lstate == IPATH_IBSTATE_INIT || lstate == IPATH_IBSTATE_ARM ||
	    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
			ipath_cdbg(SMA, "Unit %u link state %s, last "
				   "was %s\n", dd->ipath_unit,
				   ib_linkstate(lstate),
				   ib_linkstate((unsigned)
						dd->ipath_lastibcstat
						& IPATH_IBSTATE_MASK));
	}
	else {
		lstate = dd->ipath_lastibcstat & IPATH_IBSTATE_MASK;
		if (lstate == IPATH_IBSTATE_INIT ||
		    lstate == IPATH_IBSTATE_ARM ||
		    lstate == IPATH_IBSTATE_ACTIVE)
			ipath_cdbg(SMA, "Unit %u link state down"
				   " (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) {
		dd->ipath_flags |= IPATH_LINKDOWN;
		dd->ipath_flags &= ~(IPATH_LINKUNK | IPATH_LINKINIT
				     | IPATH_LINKACTIVE |
				     IPATH_LINKARMED);
		*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
		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 */
				ipath_cdbg(SMA,
					   "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);

		__ipath_layer_intr(dd, IPATH_LAYER_INT_IF_UP);
	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) {
		/*
		 * 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) {
		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,
			     unsigned supp_msgs, char msg[512])
{
	/*
	 * 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) {
		ipath_decode_err(msg, sizeof msg, dd->ipath_lasterror &
				 ~INFINIPATH_E_IBSTATUSCHANGED);
		if (dd->ipath_lasterror &
		    ~(INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL))
			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.
			 */
			ipath_dbg("Suppressed %u messages for %s\n",
				  supp_msgs, msg);
		}
	}
}

static unsigned handle_frequent_errors(struct ipath_devdata *dd,
				       ipath_err_t errs, char msg[512],
				       int *noprint)
{
	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) {
			handle_supp_msgs(dd, supp_msgs, msg);
			supp_msgs = 0;
			nmsgs = 0;
		}
	}
	else if (!nmsgs++ || time_after(nc, nextmsg_time))
		nextmsg_time = nc + HZ / 2;

	return supp_msgs;
}

static void handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
{
	char msg[512];
	u64 ignore_this_time = 0;
	int i;
	int chkerrpkts = 0, noprint = 0;
	unsigned supp_msgs;

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

	/*
	 * don't report errors that are masked (includes those always
	 * ignored)
	 */
	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);
	}

	if (!noprint && (errs & ~infinipath_e_bitsextant))
		ipath_dev_err(dd, "error interrupt with unknown errors "
			      "%llx set\n", (unsigned long long)
			      (errs & ~infinipath_e_bitsextant));

	if (errs & E_SUM_ERRS)
		ignore_this_time = handle_e_sum_errs(dd, errs);

	if (supp_msgs == 250000) {
		/*
		 * 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;
		ipath_write_kreg(dd, dd->ipath_kregs->kr_errormask,
				 ~dd->ipath_maskederrs);
		ipath_decode_err(msg, sizeof msg,
				 (dd->ipath_maskederrs & ~dd->
				  ipath_ignorederrs));

		if ((dd->ipath_maskederrs & ~dd->ipath_ignorederrs) &
		    ~(INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL))
			ipath_dev_err(dd, "Disabling error(s) %llx because "
				      "occuring too frequently (%s)\n",
				      (unsigned long long)
				      (dd->ipath_maskederrs &
				       ~dd->ipath_ignorederrs), msg);
		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.
			 */
			ipath_dbg("Disabling frequent queue full errors "
				  "(%s)\n", msg);
		}

		/*
		 * 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;