ipath_intr.c

来自「linux 内核源代码」· C语言 代码 · 共 1,209 行 · 第 1/3 页

			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);
		signal_ib_event(dd, IB_EVENT_PORT_ACTIVE);
	} else if ((val & IPATH_IBSTATE_MASK) == IPATH_IBSTATE_INIT) {
		if (dd->ipath_flags & IPATH_LINKACTIVE)
			signal_ib_event(dd, IB_EVENT_PORT_ERR);
		/*
		 * 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) {
		if (dd->ipath_flags & IPATH_LINKACTIVE)
			signal_ib_event(dd, IB_EVENT_PORT_ERR);
		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, int msgsz)
{
	/*
	 * 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) {
		int iserr;
		iserr = ipath_decode_err(msg, msgsz,
					 dd->ipath_lasterror &
					 ~INFINIPATH_E_IBSTATUSCHANGED);
		if (dd->ipath_lasterror &
			~(INFINIPATH_E_RRCVEGRFULL |
			INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS))
			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.
			 */
			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);
		}
	}
}

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

	return supp_msgs;
}

static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
{
	char msg[128];
	u64 ignore_this_time = 0;
	int i, iserr = 0;
	int chkerrpkts = 0, noprint = 0;
	unsigned supp_msgs;
	int log_idx;

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

	/* don't report errors that are masked */
	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);
	} 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);
		}
	}

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

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

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

		if (dd->ipath_maskederrs &
			~(INFINIPATH_E_RRCVEGRFULL |
			INFINIPATH_E_RRCVHDRFULL | INFINIPATH_E_PKTERRS))
			ipath_dev_err(dd, "Temporarily disabling "
			    "error(s) %llx reporting; too frequent (%s)\n",
				(unsigned long long)dd->ipath_maskederrs,
				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.
			 */
			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);
		}

		/*
		 * 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);
	}

	/* 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);
	}

	if (!errs)
		return 0;

	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;
	}
	iserr = errs & ~(E_SUM_PKTERRS | INFINIPATH_E_PKTERRS);


	/*
	 * 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++;
		for (i = 0; i < dd->ipath_cfgports; i++) {
			struct ipath_portdata *pd = dd->ipath_pd[i];
			if (i == 0) {
				hd = dd->ipath_port0head;
				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
				 */
				tl = *(u64 *) pd->port_rcvhdrtail_kvaddr;
				if (tl == dd->ipath_lastrcvhdrqtails[i])
					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;
				dd->ipath_lastrcvhdrqtails[i] = tl;
				pd->port_hdrqfull++;
				/* flush hdrqfull so that poll() sees it */
				wmb();
				wake_up_interruptible(&pd->port_wait);
			}
		}
	}
	if (errs & INFINIPATH_E_RRCVEGRFULL) {
		/*
		 * 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++;
		if (dd->ipath_port0head !=
		    (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;
	}

	if (!noprint && *msg) {
		if (iserr)
			ipath_dev_err(dd, "%s error\n", msg);
		else
			dev_info(&dd->pcidev->dev, "%s packet problems\n",
				msg);
	}
	if (dd->ipath_state_wanted & dd->ipath_flags) {
		ipath_cdbg(VERBOSE, "driver wanted state %x, iflags now %x, "
			   "waking\n", dd->ipath_state_wanted,
			   dd->ipath_flags);
		wake_up_interruptible(&ipath_state_wait);
	}

	return chkerrpkts;
}


/*
 * 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;
	__le64 val;

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

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

	/*
	 * clear all sends, because they have may been