6728711 ibnex_ioc_create_pi() and ibnex_pseudo_create_pi() shouldn't call mdi_pi_offline() in error cases

6837153 assertion failed: rc == DDI_SUCCESS, file: ../../common/os/devcfg.c, line: 4067
6841284 hermon_inter_err_chk() needs some clean-ups

5 files changed, 49 insertions, 13 deletions

diff --git a/usr/src/uts/common/io/ib/adapters/hermon/hermon.c b/usr/src/uts/common/io/ib/adapters/hermon/hermon.c
index effda2371d..6041660d65 100644
--- a/usr/src/uts/common/io/ib/adapters/hermon/hermon.c
+++ b/usr/src/uts/common/io/ib/adapters/hermon/hermon.c
@@ -56,8 +56,10 @@ int hermon_verbose = 0;
 /* Hermon HCA State Pointer */
 void *hermon_statep;
 
-int	debug_vpd = 0;
+int debug_vpd = 0;
 
+/* Disable the internal error-check polling thread */
+int hermon_no_inter_err_chk = 0;
 
 /*
  * The Hermon "userland resource database" is common to instances of the
@@ -1659,8 +1661,11 @@ hermon_hw_init(hermon_state_t *state)
 	/*
 	 * Invoke a polling thread to check the error buffer periodically.
 	 */
-	state->hs_fm_poll_thread = ddi_periodic_add(hermon_inter_err_chk,
-	    (void *)state, FM_POLL_INTERVAL, DDI_IPL_0);
+	if (!hermon_no_inter_err_chk) {
+		state->hs_fm_poll_thread = ddi_periodic_add(
+		    hermon_inter_err_chk, (void *)state, FM_POLL_INTERVAL,
+		    DDI_IPL_0);
+	}
 
 	cleanup = HERMON_DRV_CLEANUP_LEVEL5;
 
@@ -4622,6 +4627,11 @@ hermon_quiesce(dev_info_t *dip)
 	/* start fastreboot */
 	state->hs_quiescing = B_TRUE;
 
+	/* If it's in maintenance mode, do nothing but return with SUCCESS */
+	if (!HERMON_IS_OPERATIONAL(state->hs_operational_mode)) {
+		return (DDI_SUCCESS);
+	}
+
 	/* suppress Hermon FM ereports */
 	if (hermon_get_state(state) & HCA_EREPORT_FM) {
 		hermon_clr_state_nolock(state, HCA_EREPORT_FM);
diff --git a/usr/src/uts/common/io/ib/adapters/hermon/hermon_fm.c b/usr/src/uts/common/io/ib/adapters/hermon/hermon_fm.c
index 32aeebcac6..6954cd49cd 100644
--- a/usr/src/uts/common/io/ib/adapters/hermon/hermon_fm.c
+++ b/usr/src/uts/common/io/ib/adapters/hermon/hermon_fm.c
@@ -926,6 +926,10 @@ hermon_init_failure(hermon_state_t *state)
 	if (!(hermon_get_state(state) & HCA_PIO_FM))
 		return (B_FALSE);
 
+	/* check if fatal errors occur during attach */
+	if (state->hs_fm_async_fatal)
+		return (B_TRUE);
+
 	hdl = hermon_get_uarhdl(state);
 	/* Get the PIO error against UAR I/O space */
 	ddi_fm_acc_err_get(hdl, &derr, DDI_FME_VERSION);
@@ -1448,7 +1452,14 @@ hermon_inter_err_chk(void *arg)
 
 	if (word != 0) {
 		HERMON_FMANOTE(state, HERMON_FMA_INTERNAL);
-		hermon_fm_ereport(state, HCA_IBA_ERR, HCA_ERR_FATAL);
+		/* if fm_disable is on, Hermon FM functions don't work */
+		if (state->hs_fm_disable) {
+			cmn_err(CE_PANIC,
+			    "Hermon Fatal Internal Error. "
+			    "Hermon state=0x%p", (void *)state);
+		} else {
+			hermon_fm_ereport(state, HCA_IBA_ERR, HCA_ERR_FATAL);
+		}
 	}
 
 	/* issue the ereport pended in the interrupt context */
diff --git a/usr/src/uts/common/io/ib/adapters/tavor/tavor.c b/usr/src/uts/common/io/ib/adapters/tavor/tavor.c
index 5c6570d194..9c29ddb581 100644
--- a/usr/src/uts/common/io/ib/adapters/tavor/tavor.c
+++ b/usr/src/uts/common/io/ib/adapters/tavor/tavor.c
@@ -3489,6 +3489,11 @@ tavor_quiesce(dev_info_t *dip)
 	/* start fastreboot */
 	state->ts_quiescing = B_TRUE;
 
+	/* If it's in maintenance mode, do nothing but return with SUCCESS */
+	if (!TAVOR_IS_OPERATIONAL(state->ts_operational_mode)) {
+		return (DDI_SUCCESS);
+	}
+
 	/* Shutdown HCA ports */
 	if (tavor_hca_ports_shutdown(state,
 	    state->ts_cfg_profile->cp_num_ports) != TAVOR_CMD_SUCCESS) {
diff --git a/usr/src/uts/common/io/ib/ibnex/ibnex.c b/usr/src/uts/common/io/ib/ibnex/ibnex.c
index 88c03b4f09..c0e3b942e1 100644
--- a/usr/src/uts/common/io/ib/ibnex/ibnex.c
+++ b/usr/src/uts/common/io/ib/ibnex/ibnex.c
@@ -1735,14 +1735,14 @@ ibnex_bus_unconfig(dev_info_t *parent,
 			    ndp; ndp = ndp->node_next) {
 				dip = ndp->node_dip;
 				if (dip && (ddi_driver_major(dip) == major)) {
-					ibnex_offline_childdip(dip);
+					(void) ibnex_offline_childdip(dip);
 				}
 			}
 			for (ndp = ibnex.ibnex_pseudo_node_head;
 			    ndp; ndp = ndp->node_next) {
 				dip = ndp->node_dip;
 				if (dip && (ddi_driver_major(dip) == major)) {
-					ibnex_offline_childdip(dip);
+					(void) ibnex_offline_childdip(dip);
 				}
 			}
 		}
diff --git a/usr/src/uts/common/io/ib/ibnex/ibnex_ioctl.c b/usr/src/uts/common/io/ib/ibnex/ibnex_ioctl.c
index 5aecb859a1..85e7386e83 100644
--- a/usr/src/uts/common/io/ib/ibnex/ibnex_ioctl.c
+++ b/usr/src/uts/common/io/ib/ibnex/ibnex_ioctl.c
@@ -2625,7 +2625,7 @@ ibnex_ioc_fininode(dev_info_t *dip, ibnex_ioc_node_t *ioc_nodep)
 int
 ibnex_offline_childdip(dev_info_t *dip)
 {
-	int		rval = MDI_SUCCESS;
+	int		rval = MDI_SUCCESS, rval2;
 	mdi_pathinfo_t	*path = NULL, *temp;
 
 	IBTF_DPRINTF_L4("ibnex", "\toffline_childdip; begin");
@@ -2637,15 +2637,25 @@ ibnex_offline_childdip(dev_info_t *dip)
 	for (path = mdi_get_next_phci_path(dip, path); path; ) {
 		IBTF_DPRINTF_L4("ibnex", "\toffline_childdip: "
 		    "offling path %p", path);
-		rval = mdi_pi_offline(path, NDI_UNCONFIG);
-		if (rval != MDI_SUCCESS) {
-			IBTF_DPRINTF_L2("ibnex", "\toffline_childdip: "
-			    "mdi_pi_offline failed %p", dip);
-			break;
+		rval2 = MDI_SUCCESS;
+		if (MDI_PI_IS_ONLINE(path)) {
+			rval2 = mdi_pi_offline(path, NDI_UNCONFIG);
+			/* If it cannot be offlined, log this path and error */
+			if (rval2 != MDI_SUCCESS) {
+				rval = rval2;
+				cmn_err(CE_WARN,
+				    "!ibnex\toffline_childdip (0x%p): "
+				    "mdi_pi_offline path (0x%p) failed with %d",
+				    (void *)dip, (void *)path, rval2);
+			}
 		}
+		/* prepare the next path */
 		temp = path;
 		path = mdi_get_next_phci_path(dip, path);
-		(void) mdi_pi_free(temp, 0);
+		/* free the offline path */
+		if (rval2 == MDI_SUCCESS) {
+			(void) mdi_pi_free(temp, 0);
+		}
 	}
 	return (rval);
 }