1 files changed, 473 insertions, 365 deletions

diff --git a/drivers/infiniband/hw/ehca/ehca_irq.c b/drivers/infiniband/hw/ehca/ehca_irq.c
index 2a65b5be197..8615d7cf7e0 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.c
+++ b/drivers/infiniband/hw/ehca/ehca_irq.c
@@ -5,6 +5,8 @@
  *
  *  Authors: Heiko J Schick <schickhj@de.ibm.com>
  *           Khadija Souissi <souissi@de.ibm.com>
+ *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
+ *           Joachim Fenkes <fenkes@de.ibm.com>
  *
  *  Copyright (c) 2005 IBM Corporation
  *
@@ -39,37 +41,38 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+#include <linux/smpboot.h>
+
 #include "ehca_classes.h"
 #include "ehca_irq.h"
 #include "ehca_iverbs.h"
 #include "ehca_tools.h"
 #include "hcp_if.h"
 #include "hipz_fns.h"
-
-#define EQE_COMPLETION_EVENT   EHCA_BMASK_IBM(1,1)
-#define EQE_CQ_QP_NUMBER       EHCA_BMASK_IBM(8,31)
-#define EQE_EE_IDENTIFIER      EHCA_BMASK_IBM(2,7)
-#define EQE_CQ_NUMBER          EHCA_BMASK_IBM(8,31)
-#define EQE_QP_NUMBER          EHCA_BMASK_IBM(8,31)
-#define EQE_QP_TOKEN           EHCA_BMASK_IBM(32,63)
-#define EQE_CQ_TOKEN           EHCA_BMASK_IBM(32,63)
-
-#define NEQE_COMPLETION_EVENT  EHCA_BMASK_IBM(1,1)
-#define NEQE_EVENT_CODE        EHCA_BMASK_IBM(2,7)
-#define NEQE_PORT_NUMBER       EHCA_BMASK_IBM(8,15)
-#define NEQE_PORT_AVAILABILITY EHCA_BMASK_IBM(16,16)
-
-#define ERROR_DATA_LENGTH      EHCA_BMASK_IBM(52,63)
-#define ERROR_DATA_TYPE        EHCA_BMASK_IBM(0,7)
-
-#ifdef CONFIG_INFINIBAND_EHCA_SCALING
+#include "ipz_pt_fn.h"
+
+#define EQE_COMPLETION_EVENT   EHCA_BMASK_IBM( 1,  1)
+#define EQE_CQ_QP_NUMBER       EHCA_BMASK_IBM( 8, 31)
+#define EQE_EE_IDENTIFIER      EHCA_BMASK_IBM( 2,  7)
+#define EQE_CQ_NUMBER          EHCA_BMASK_IBM( 8, 31)
+#define EQE_QP_NUMBER          EHCA_BMASK_IBM( 8, 31)
+#define EQE_QP_TOKEN           EHCA_BMASK_IBM(32, 63)
+#define EQE_CQ_TOKEN           EHCA_BMASK_IBM(32, 63)
+
+#define NEQE_COMPLETION_EVENT  EHCA_BMASK_IBM( 1,  1)
+#define NEQE_EVENT_CODE        EHCA_BMASK_IBM( 2,  7)
+#define NEQE_PORT_NUMBER       EHCA_BMASK_IBM( 8, 15)
+#define NEQE_PORT_AVAILABILITY EHCA_BMASK_IBM(16, 16)
+#define NEQE_DISRUPTIVE        EHCA_BMASK_IBM(16, 16)
+#define NEQE_SPECIFIC_EVENT    EHCA_BMASK_IBM(16, 23)
+
+#define ERROR_DATA_LENGTH      EHCA_BMASK_IBM(52, 63)
+#define ERROR_DATA_TYPE        EHCA_BMASK_IBM( 0,  7)
 
 static void queue_comp_task(struct ehca_cq *__cq);
 
-static struct ehca_comp_pool* pool;
-static struct notifier_block comp_pool_callback_nb;
-
-#endif
+static struct ehca_comp_pool *pool;
 
 static inline void comp_event_callback(struct ehca_cq *cq)
 {
@@ -83,8 +86,8 @@ static inline void comp_event_callback(struct ehca_cq *cq)
 	return;
 }
 
-static void print_error_data(struct ehca_shca * shca, void* data,
-			     u64* rblock, int length)
+static void print_error_data(struct ehca_shca *shca, void *data,
+			     u64 *rblock, int length)
 {
 	u64 type = EHCA_BMASK_GET(ERROR_DATA_TYPE, rblock[2]);
 	u64 resource = rblock[1];
@@ -92,37 +95,37 @@ static void print_error_data(struct ehca_shca * shca, void* data,
 	switch (type) {
 	case 0x1: /* Queue Pair */
 	{
-		struct ehca_qp *qp = (struct ehca_qp*)data;
+		struct ehca_qp *qp = (struct ehca_qp *)data;
 
 		/* only print error data if AER is set */
 		if (rblock[6] == 0)
 			return;
 
 		ehca_err(&shca->ib_device,
-			 "QP 0x%x (resource=%lx) has errors.",
+			 "QP 0x%x (resource=%llx) has errors.",
 			 qp->ib_qp.qp_num, resource);
 		break;
 	}
 	case 0x4: /* Completion Queue */
 	{
-		struct ehca_cq *cq = (struct ehca_cq*)data;
+		struct ehca_cq *cq = (struct ehca_cq *)data;
 
 		ehca_err(&shca->ib_device,
-			 "CQ 0x%x (resource=%lx) has errors.",
+			 "CQ 0x%x (resource=%llx) has errors.",
 			 cq->cq_number, resource);
 		break;
 	}
 	default:
 		ehca_err(&shca->ib_device,
-			 "Unknown errror type: %lx on %s.",
+			 "Unknown error type: %llx on %s.",
 			 type, shca->ib_device.name);
 		break;
 	}
 
-	ehca_err(&shca->ib_device, "Error data is available: %lx.", resource);
+	ehca_err(&shca->ib_device, "Error data is available: %llx.", resource);
 	ehca_err(&shca->ib_device, "EHCA ----- error data begin "
 		 "---------------------------------------------------");
-	ehca_dmp(rblock, length, "resource=%lx", resource);
+	ehca_dmp(rblock, length, "resource=%llx", resource);
 	ehca_err(&shca->ib_device, "EHCA ----- error data end "
 		 "----------------------------------------------------");
 
@@ -137,91 +140,123 @@ int ehca_error_data(struct ehca_shca *shca, void *data,
 	u64 *rblock;
 	unsigned long block_count;
 
-	rblock = kzalloc(H_CB_ALIGNMENT, GFP_KERNEL);
+	rblock = ehca_alloc_fw_ctrlblock(GFP_ATOMIC);
 	if (!rblock) {
 		ehca_err(&shca->ib_device, "Cannot allocate rblock memory.");
 		ret = -ENOMEM;
 		goto error_data1;
 	}
 
+	/* rblock must be 4K aligned and should be 4K large */
 	ret = hipz_h_error_data(shca->ipz_hca_handle,
 				resource,
 				rblock,
 				&block_count);
 
-	if (ret == H_R_STATE) {
+	if (ret == H_R_STATE)
 		ehca_err(&shca->ib_device,
-			 "No error data is available: %lx.", resource);
-	}
+			 "No error data is available: %llx.", resource);
 	else if (ret == H_SUCCESS) {
 		int length;
 
 		length = EHCA_BMASK_GET(ERROR_DATA_LENGTH, rblock[0]);
 
-		if (length > PAGE_SIZE)
-			length = PAGE_SIZE;
+		if (length > EHCA_PAGESIZE)
+			length = EHCA_PAGESIZE;
 
 		print_error_data(shca, data, rblock, length);
-	}
-	else {
+	} else
 		ehca_err(&shca->ib_device,
-			 "Error data could not be fetched: %lx", resource);
-	}
+			 "Error data could not be fetched: %llx", resource);
 
-	kfree(rblock);
+	ehca_free_fw_ctrlblock(rblock);
 
 error_data1:
 	return ret;
 
 }
 
-static void qp_event_callback(struct ehca_shca *shca,
-			      u64 eqe,
+static void dispatch_qp_event(struct ehca_shca *shca, struct ehca_qp *qp,
 			      enum ib_event_type event_type)
 {
 	struct ib_event event;
+
+	/* PATH_MIG without the QP ever having been armed is false alarm */
+	if (event_type == IB_EVENT_PATH_MIG && !qp->mig_armed)
+		return;
+
+	event.device = &shca->ib_device;
+	event.event = event_type;
+
+	if (qp->ext_type == EQPT_SRQ) {
+		if (!qp->ib_srq.event_handler)
+			return;
+
+		event.element.srq = &qp->ib_srq;
+		qp->ib_srq.event_handler(&event, qp->ib_srq.srq_context);
+	} else {
+		if (!qp->ib_qp.event_handler)
+			return;
+
+		event.element.qp = &qp->ib_qp;
+		qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
+	}
+}
+
+static void qp_event_callback(struct ehca_shca *shca, u64 eqe,
+			      enum ib_event_type event_type, int fatal)
+{
 	struct ehca_qp *qp;
-	unsigned long flags;
 	u32 token = EHCA_BMASK_GET(EQE_QP_TOKEN, eqe);
 
-	spin_lock_irqsave(&ehca_qp_idr_lock, flags);
+	read_lock(&ehca_qp_idr_lock);
 	qp = idr_find(&ehca_qp_idr, token);
-	spin_unlock_irqrestore(&ehca_qp_idr_lock, flags);
-
+	if (qp)
+		atomic_inc(&qp->nr_events);
+	read_unlock(&ehca_qp_idr_lock);
 
 	if (!qp)
 		return;
 
-	ehca_error_data(shca, qp, qp->ipz_qp_handle.handle);
-
-	if (!qp->ib_qp.event_handler)
-		return;
+	if (fatal)
+		ehca_error_data(shca, qp, qp->ipz_qp_handle.handle);
 
-	event.device     = &shca->ib_device;
-	event.event      = event_type;
-	event.element.qp = &qp->ib_qp;
+	dispatch_qp_event(shca, qp, fatal && qp->ext_type == EQPT_SRQ ?
+			  IB_EVENT_SRQ_ERR : event_type);
 
-	qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
+	/*
+	 * eHCA only processes one WQE at a time for SRQ base QPs,
+	 * so the last WQE has been processed as soon as the QP enters
+	 * error state.
+	 */
+	if (fatal && qp->ext_type == EQPT_SRQBASE)
+		dispatch_qp_event(shca, qp, IB_EVENT_QP_LAST_WQE_REACHED);
 
+	if (atomic_dec_and_test(&qp->nr_events))
+		wake_up(&qp->wait_completion);
 	return;
 }
 
 static void cq_event_callback(struct ehca_shca *shca,
-					  u64 eqe)
+			      u64 eqe)
 {
 	struct ehca_cq *cq;
-	unsigned long flags;
 	u32 token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe);
 
-	spin_lock_irqsave(&ehca_cq_idr_lock, flags);
+	read_lock(&ehca_cq_idr_lock);
 	cq = idr_find(&ehca_cq_idr, token);
-	spin_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+	if (cq)
+		atomic_inc(&cq->nr_events);
+	read_unlock(&ehca_cq_idr_lock);
 
 	if (!cq)
 		return;
 
 	ehca_error_data(shca, cq, cq->ipz_cq_handle.handle);
 
+	if (atomic_dec_and_test(&cq->nr_events))
+		wake_up(&cq->wait_completion);
+
 	return;
 }
 
@@ -231,17 +266,17 @@ static void parse_identifier(struct ehca_shca *shca, u64 eqe)
 
 	switch (identifier) {
 	case 0x02: /* path migrated */
-		qp_event_callback(shca, eqe, IB_EVENT_PATH_MIG);
+		qp_event_callback(shca, eqe, IB_EVENT_PATH_MIG, 0);
 		break;
 	case 0x03: /* communication established */
-		qp_event_callback(shca, eqe, IB_EVENT_COMM_EST);
+		qp_event_callback(shca, eqe, IB_EVENT_COMM_EST, 0);
 		break;
 	case 0x04: /* send queue drained */
-		qp_event_callback(shca, eqe, IB_EVENT_SQ_DRAINED);
+		qp_event_callback(shca, eqe, IB_EVENT_SQ_DRAINED, 0);
 		break;
 	case 0x05: /* QP error */
 	case 0x06: /* QP error */
-		qp_event_callback(shca, eqe, IB_EVENT_QP_FATAL);
+		qp_event_callback(shca, eqe, IB_EVENT_QP_FATAL, 1);
 		break;
 	case 0x07: /* CQ error */
 	case 0x08: /* CQ error */
@@ -268,13 +303,18 @@ static void parse_identifier(struct ehca_shca *shca, u64 eqe)
 	case 0x11: /* unaffiliated access error */
 		ehca_err(&shca->ib_device, "Unaffiliated access error.");
 		break;
-	case 0x12: /* path migrating error */
-		ehca_err(&shca->ib_device, "Path migration error.");
+	case 0x12: /* path migrating */
+		ehca_err(&shca->ib_device, "Path migrating.");
 		break;
 	case 0x13: /* interface trace stopped */
 		ehca_err(&shca->ib_device, "Interface trace stopped.");
 		break;
 	case 0x14: /* first error capture info available */
+		ehca_info(&shca->ib_device, "First error capture available");
+		break;
+	case 0x15: /* SRQ limit reached */
+		qp_event_callback(shca, eqe, IB_EVENT_SRQ_LIMIT_REACHED, 0);
+		break;
 	default:
 		ehca_err(&shca->ib_device, "Unknown identifier: %x on %s.",
 			 identifier, shca->ib_device.name);
@@ -284,30 +324,90 @@ static void parse_identifier(struct ehca_shca *shca, u64 eqe)
 	return;
 }
 
-static void parse_ec(struct ehca_shca *shca, u64 eqe)
+static void dispatch_port_event(struct ehca_shca *shca, int port_num,
+				enum ib_event_type type, const char *msg)
 {
 	struct ib_event event;
+
+	ehca_info(&shca->ib_device, "port %d %s.", port_num, msg);
+	event.device = &shca->ib_device;
+	event.event = type;
+	event.element.port_num = port_num;
+	ib_dispatch_event(&event);
+}
+
+static void notify_port_conf_change(struct ehca_shca *shca, int port_num)
+{
+	struct ehca_sma_attr  new_attr;
+	struct ehca_sma_attr *old_attr = &shca->sport[port_num - 1].saved_attr;
+
+	ehca_query_sma_attr(shca, port_num, &new_attr);
+
+	if (new_attr.sm_sl  != old_attr->sm_sl ||
+	    new_attr.sm_lid != old_attr->sm_lid)
+		dispatch_port_event(shca, port_num, IB_EVENT_SM_CHANGE,
+				    "SM changed");
+
+	if (new_attr.lid != old_attr->lid ||
+	    new_attr.lmc != old_attr->lmc)
+		dispatch_port_event(shca, port_num, IB_EVENT_LID_CHANGE,
+				    "LID changed");
+
+	if (new_attr.pkey_tbl_len != old_attr->pkey_tbl_len ||
+	    memcmp(new_attr.pkeys, old_attr->pkeys,
+		   sizeof(u16) * new_attr.pkey_tbl_len))
+		dispatch_port_event(shca, port_num, IB_EVENT_PKEY_CHANGE,
+				    "P_Key changed");
+
+	*old_attr = new_attr;
+}
+
+/* replay modify_qp for sqps -- return 0 if all is well, 1 if AQP1 destroyed */
+static int replay_modify_qp(struct ehca_sport *sport)
+{
+	int aqp1_destroyed;
+	unsigned long flags;
+
+	spin_lock_irqsave(&sport->mod_sqp_lock, flags);
+
+	aqp1_destroyed = !sport->ibqp_sqp[IB_QPT_GSI];
+
+	if (sport->ibqp_sqp[IB_QPT_SMI])
+		ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_SMI]);
+	if (!aqp1_destroyed)
+		ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_GSI]);
+
+	spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
+
+	return aqp1_destroyed;
+}
+
+static void parse_ec(struct ehca_shca *shca, u64 eqe)
+{
 	u8 ec   = EHCA_BMASK_GET(NEQE_EVENT_CODE, eqe);
 	u8 port = EHCA_BMASK_GET(NEQE_PORT_NUMBER, eqe);
+	u8 spec_event;
+	struct ehca_sport *sport = &shca->sport[port - 1];
 
 	switch (ec) {
 	case 0x30: /* port availability change */
 		if (EHCA_BMASK_GET(NEQE_PORT_AVAILABILITY, eqe)) {
-			ehca_info(&shca->ib_device,
-				  "port %x is active.", port);
-			event.device = &shca->ib_device;
-			event.event = IB_EVENT_PORT_ACTIVE;
-			event.element.port_num = port;
-			shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
-			ib_dispatch_event(&event);
+			/* only replay modify_qp calls in autodetect mode;
+			 * if AQP1 was destroyed, the port is already down
+			 * again and we can drop the event.
+			 */
+			if (ehca_nr_ports < 0)
+				if (replay_modify_qp(sport))
+					break;
+
+			sport->port_state = IB_PORT_ACTIVE;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
+					    "is active");
+			ehca_query_sma_attr(shca, port, &sport->saved_attr);
 		} else {
-			ehca_info(&shca->ib_device,
-				  "port %x is inactive.", port);
-			event.device = &shca->ib_device;
-			event.event = IB_EVENT_PORT_ERR;
-			event.element.port_num = port;
-			shca->sport[port - 1].port_state = IB_PORT_DOWN;
-			ib_dispatch_event(&event);
+			sport->port_state = IB_PORT_DOWN;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ERR,
+					    "is inactive");
 		}
 		break;
 	case 0x31:
@@ -315,24 +415,21 @@ static void parse_ec(struct ehca_shca *shca, u64 eqe)
 		 * disruptive change is caused by
 		 * LID, PKEY or SM change
 		 */
-		ehca_warn(&shca->ib_device,
-			  "disruptive port %x configuration change", port);
-
-		ehca_info(&shca->ib_device,
-			 "port %x is inactive.", port);
-		event.device = &shca->ib_device;
-		event.event = IB_EVENT_PORT_ERR;
-		event.element.port_num = port;
-		shca->sport[port - 1].port_state = IB_PORT_DOWN;
-		ib_dispatch_event(&event);
-
-		ehca_info(&shca->ib_device,
-			 "port %x is active.", port);
-		event.device = &shca->ib_device;
-		event.event = IB_EVENT_PORT_ACTIVE;
-		event.element.port_num = port;
-		shca->sport[port - 1].port_state = IB_PORT_ACTIVE;
-		ib_dispatch_event(&event);
+		if (EHCA_BMASK_GET(NEQE_DISRUPTIVE, eqe)) {
+			ehca_warn(&shca->ib_device, "disruptive port "
+				  "%d configuration change", port);
+
+			sport->port_state = IB_PORT_DOWN;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ERR,
+					    "is inactive");
+
+			sport->port_state = IB_PORT_ACTIVE;
+			dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
+					    "is active");
+			ehca_query_sma_attr(shca, port,
+					    &sport->saved_attr);
+		} else
+			notify_port_conf_change(shca, port);
 		break;
 	case 0x32: /* adapter malfunction */
 		ehca_err(&shca->ib_device, "Adapter malfunction.");
@@ -340,6 +437,16 @@ static void parse_ec(struct ehca_shca *shca, u64 eqe)
 	case 0x33:  /* trace stopped */
 		ehca_err(&shca->ib_device, "Traced stopped.");
 		break;
+	case 0x34: /* util async event */
+		spec_event = EHCA_BMASK_GET(NEQE_SPECIFIC_EVENT, eqe);
+		if (spec_event == 0x80) /* client reregister required */
+			dispatch_port_event(shca, port,
+					    IB_EVENT_CLIENT_REREGISTER,
+					    "client reregister req.");
+		else
+			ehca_warn(&shca->ib_device, "Unknown util async "
+				  "event %x on port %x", spec_event, port);
+		break;
 	default:
 		ehca_err(&shca->ib_device, "Unknown event code: %x on %s.",
 			 ec, shca->ib_device.name);
@@ -360,7 +467,7 @@ static inline void reset_eq_pending(struct ehca_cq *cq)
 	return;
 }
 
-irqreturn_t ehca_interrupt_neq(int irq, void *dev_id, struct pt_regs *regs)
+irqreturn_t ehca_interrupt_neq(int irq, void *dev_id)
 {
 	struct ehca_shca *shca = (struct ehca_shca*)dev_id;
 
@@ -375,13 +482,13 @@ void ehca_tasklet_neq(unsigned long data)
 	struct ehca_eqe *eqe;
 	u64 ret;
 
-	eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
+	eqe = ehca_poll_eq(shca, &shca->neq);
 
 	while (eqe) {
 		if (!EHCA_BMASK_GET(NEQE_COMPLETION_EVENT, eqe->entry))
 			parse_ec(shca, eqe->entry);
 
-		eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
+		eqe = ehca_poll_eq(shca, &shca->neq);
 	}
 
 	ret = hipz_h_reset_event(shca->ipz_hca_handle,
@@ -393,7 +500,7 @@ void ehca_tasklet_neq(unsigned long data)
 	return;
 }
 
-irqreturn_t ehca_interrupt_eq(int irq, void *dev_id, struct pt_regs *regs)
+irqreturn_t ehca_interrupt_eq(int irq, void *dev_id)
 {
 	struct ehca_shca *shca = (struct ehca_shca*)dev_id;
 
@@ -402,361 +509,362 @@ irqreturn_t ehca_interrupt_eq(int irq, void *dev_id, struct pt_regs *regs)
 	return IRQ_HANDLED;
 }
 
-void ehca_tasklet_eq(unsigned long data)
+
+static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 {
-	struct ehca_shca *shca = (struct ehca_shca*)data;
-	struct ehca_eqe *eqe;
-	int int_state;
-	int query_cnt = 0;
+	u64 eqe_value;
+	u32 token;
+	struct ehca_cq *cq;
 
-	do {
-		eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->eq);
+	eqe_value = eqe->entry;
+	ehca_dbg(&shca->ib_device, "eqe_value=%llx", eqe_value);
+	if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
+		ehca_dbg(&shca->ib_device, "Got completion event");
+		token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
+		read_lock(&ehca_cq_idr_lock);
+		cq = idr_find(&ehca_cq_idr, token);
+		if (cq)
+			atomic_inc(&cq->nr_events);
+		read_unlock(&ehca_cq_idr_lock);
+		if (cq == NULL) {
+			ehca_err(&shca->ib_device,
+				 "Invalid eqe for non-existing cq token=%x",
+				 token);
+			return;
+		}
+		reset_eq_pending(cq);
+		if (ehca_scaling_code)
+			queue_comp_task(cq);
+		else {
+			comp_event_callback(cq);
+			if (atomic_dec_and_test(&cq->nr_events))
+				wake_up(&cq->wait_completion);
+		}
+	} else {
+		ehca_dbg(&shca->ib_device, "Got non completion event");
+		parse_identifier(shca, eqe_value);
+	}
+}
 
-		if ((shca->hw_level >= 2) && eqe)
-			int_state = 1;
-		else
-			int_state = 0;
-
-		while ((int_state == 1) || eqe) {
-			while (eqe) {
-				u64 eqe_value = eqe->entry;
-
-				ehca_dbg(&shca->ib_device,
-					 "eqe_value=%lx", eqe_value);
-
-				/* TODO: better structure */
-				if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT,
-						   eqe_value)) {
-					unsigned long flags;
-					u32 token;
-					struct ehca_cq *cq;
-
-					ehca_dbg(&shca->ib_device,
-						 "... completion event");
-					token =
-						EHCA_BMASK_GET(EQE_CQ_TOKEN,
-							       eqe_value);
-					spin_lock_irqsave(&ehca_cq_idr_lock,
-							  flags);
-					cq = idr_find(&ehca_cq_idr, token);
-
-					if (cq == NULL) {
-						spin_unlock(&ehca_cq_idr_lock);
-						break;
-					}
-
-					reset_eq_pending(cq);
-#ifdef CONFIG_INFINIBAND_EHCA_SCALING
-					queue_comp_task(cq);
-					spin_unlock_irqrestore(&ehca_cq_idr_lock,
-							       flags);
-#else
-					spin_unlock_irqrestore(&ehca_cq_idr_lock,
-							       flags);
-					comp_event_callback(cq);
-#endif
-				} else {
-					ehca_dbg(&shca->ib_device,
-						 "... non completion event");
-					parse_identifier(shca, eqe_value);
-				}
-				eqe =
-					(struct ehca_eqe *)ehca_poll_eq(shca,
-								    &shca->eq);
-			}
+void ehca_process_eq(struct ehca_shca *shca, int is_irq)
+{
+	struct ehca_eq *eq = &shca->eq;
+	struct ehca_eqe_cache_entry *eqe_cache = eq->eqe_cache;
+	u64 eqe_value, ret;
+	int eqe_cnt, i;
+	int eq_empty = 0;
+
+	spin_lock(&eq->irq_spinlock);
+	if (is_irq) {
+		const int max_query_cnt = 100;
+		int query_cnt = 0;
+		int int_state = 1;
+		do {
+			int_state = hipz_h_query_int_state(
+				shca->ipz_hca_handle, eq->ist);
+			query_cnt++;
+			iosync();
+		} while (int_state && query_cnt < max_query_cnt);
+		if (unlikely((query_cnt == max_query_cnt)))
+			ehca_dbg(&shca->ib_device, "int_state=%x query_cnt=%x",
+				 int_state, query_cnt);
+	}
 
-			if (shca->hw_level >= 2) {
-				int_state =
-				    hipz_h_query_int_state(shca->ipz_hca_handle,
-							   shca->eq.ist);
-				query_cnt++;
-				iosync();
-				if (query_cnt >= 100) {
-					query_cnt = 0;
-					int_state = 0;
-				}
+	/* read out all eqes */
+	eqe_cnt = 0;
+	do {
+		u32 token;
+		eqe_cache[eqe_cnt].eqe = ehca_poll_eq(shca, eq);
+		if (!eqe_cache[eqe_cnt].eqe)
+			break;
+		eqe_value = eqe_cache[eqe_cnt].eqe->entry;
+		if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
+			token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
+			read_lock(&ehca_cq_idr_lock);
+			eqe_cache[eqe_cnt].cq = idr_find(&ehca_cq_idr, token);
+			if (eqe_cache[eqe_cnt].cq)
+				atomic_inc(&eqe_cache[eqe_cnt].cq->nr_events);
+			read_unlock(&ehca_cq_idr_lock);
+			if (!eqe_cache[eqe_cnt].cq) {
+				ehca_err(&shca->ib_device,
+					 "Invalid eqe for non-existing cq "
+					 "token=%x", token);
+				continue;
 			}
-			eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->eq);
-
+		} else
+			eqe_cache[eqe_cnt].cq = NULL;
+		eqe_cnt++;
+	} while (eqe_cnt < EHCA_EQE_CACHE_SIZE);
+	if (!eqe_cnt) {
+		if (is_irq)
+			ehca_dbg(&shca->ib_device,
+				 "No eqe found for irq event");
+		goto unlock_irq_spinlock;
+	} else if (!is_irq) {
+		ret = hipz_h_eoi(eq->ist);
+		if (ret != H_SUCCESS)
+			ehca_err(&shca->ib_device,
+				 "bad return code EOI -rc = %lld\n", ret);
+		ehca_dbg(&shca->ib_device, "deadman found %x eqe", eqe_cnt);
+	}
+	if (unlikely(eqe_cnt == EHCA_EQE_CACHE_SIZE))
+		ehca_dbg(&shca->ib_device, "too many eqes for one irq event");
+	/* enable irq for new packets */
+	for (i = 0; i < eqe_cnt; i++) {
+		if (eq->eqe_cache[i].cq)
+			reset_eq_pending(eq->eqe_cache[i].cq);
+	}
+	/* check eq */
+	spin_lock(&eq->spinlock);
+	eq_empty = (!ipz_eqit_eq_peek_valid(&shca->eq.ipz_queue));
+	spin_unlock(&eq->spinlock);
+	/* call completion handler for cached eqes */
+	for (i = 0; i < eqe_cnt; i++)
+		if (eq->eqe_cache[i].cq) {
+			if (ehca_scaling_code)
+				queue_comp_task(eq->eqe_cache[i].cq);
+			else {
+				struct ehca_cq *cq = eq->eqe_cache[i].cq;
+				comp_event_callback(cq);
+				if (atomic_dec_and_test(&cq->nr_events))
+					wake_up(&cq->wait_completion);
+			}
+		} else {
+			ehca_dbg(&shca->ib_device, "Got non completion event");
+			parse_identifier(shca, eq->eqe_cache[i].eqe->entry);
 		}
-	} while (int_state != 0);
-
-	return;
+	/* poll eq if not empty */
+	if (eq_empty)
+		goto unlock_irq_spinlock;
+	do {
+		struct ehca_eqe *eqe;
+		eqe = ehca_poll_eq(shca, &shca->eq);
+		if (!eqe)
+			break;
+		process_eqe(shca, eqe);
+	} while (1);
+
+unlock_irq_spinlock:
+	spin_unlock(&eq->irq_spinlock);
 }
 
-#ifdef CONFIG_INFINIBAND_EHCA_SCALING
-
-static inline int find_next_online_cpu(struct ehca_comp_pool* pool)
+void ehca_tasklet_eq(unsigned long data)
 {
-	unsigned long flags_last_cpu;
+	ehca_process_eq((struct ehca_shca*)data, 1);
+}
 
-	if (ehca_debug_level)
-		ehca_dmp(&cpu_online_map, sizeof(cpumask_t), "");
+static int find_next_online_cpu(struct ehca_comp_pool *pool)
+{
+	int cpu;
+	unsigned long flags;
 
-	spin_lock_irqsave(&pool->last_cpu_lock, flags_last_cpu);
-	pool->last_cpu = next_cpu(pool->last_cpu, cpu_online_map);
-	if (pool->last_cpu == NR_CPUS)
-		pool->last_cpu = first_cpu(cpu_online_map);
-	spin_unlock_irqrestore(&pool->last_cpu_lock, flags_last_cpu);
+	WARN_ON_ONCE(!in_interrupt());
+	if (ehca_debug_level >= 3)
+		ehca_dmp(cpu_online_mask, cpumask_size(), "");
 
-	return pool->last_cpu;
+	spin_lock_irqsave(&pool->last_cpu_lock, flags);
+	do {
+		cpu = cpumask_next(pool->last_cpu, cpu_online_mask);
+		if (cpu >= nr_cpu_ids)
+			cpu = cpumask_first(cpu_online_mask);
+		pool->last_cpu = cpu;
+	} while (!per_cpu_ptr(pool->cpu_comp_tasks, cpu)->active);
+	spin_unlock_irqrestore(&pool->last_cpu_lock, flags);
+
+	return cpu;
 }
 
 static void __queue_comp_task(struct ehca_cq *__cq,
-			      struct ehca_cpu_comp_task *cct)
+			      struct ehca_cpu_comp_task *cct,
+			      struct task_struct *thread)
 {
-	unsigned long flags_cct;
-	unsigned long flags_cq;
+	unsigned long flags;
 
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
-	spin_lock_irqsave(&__cq->task_lock, flags_cq);
+	spin_lock_irqsave(&cct->task_lock, flags);
+	spin_lock(&__cq->task_lock);
 
 	if (__cq->nr_callbacks == 0) {
 		__cq->nr_callbacks++;
 		list_add_tail(&__cq->entry, &cct->cq_list);
 		cct->cq_jobs++;
-		wake_up(&cct->wait_queue);
-	}
-	else
+		wake_up_process(thread);
+	} else
 		__cq->nr_callbacks++;
 
-	spin_unlock_irqrestore(&__cq->task_lock, flags_cq);
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
+	spin_unlock(&__cq->task_lock);
+	spin_unlock_irqrestore(&cct->task_lock, flags);
 }
 
 static void queue_comp_task(struct ehca_cq *__cq)
 {
-	int cpu;
 	int cpu_id;
 	struct ehca_cpu_comp_task *cct;
+	struct task_struct *thread;
+	int cq_jobs;
+	unsigned long flags;
 
-	cpu = get_cpu();
 	cpu_id = find_next_online_cpu(pool);
-
 	BUG_ON(!cpu_online(cpu_id));
 
 	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
+	thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu_id);
+	BUG_ON(!cct || !thread);
 
-	if (cct->cq_jobs > 0) {
+	spin_lock_irqsave(&cct->task_lock, flags);
+	cq_jobs = cct->cq_jobs;
+	spin_unlock_irqrestore(&cct->task_lock, flags);
+	if (cq_jobs > 0) {
 		cpu_id = find_next_online_cpu(pool);
 		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
+		thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu_id);
+		BUG_ON(!cct || !thread);
 	}
-
-	__queue_comp_task(__cq, cct);
-
-	put_cpu();
-
-	return;
+	__queue_comp_task(__cq, cct, thread);
 }
 
-static void run_comp_task(struct ehca_cpu_comp_task* cct)
+static void run_comp_task(struct ehca_cpu_comp_task *cct)
 {
 	struct ehca_cq *cq;
-	unsigned long flags_cct;
-	unsigned long flags_cq;
-
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
 
 	while (!list_empty(&cct->cq_list)) {
 		cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
-		spin_unlock_irqrestore(&cct->task_lock, flags_cct);
+		spin_unlock_irq(&cct->task_lock);
+
 		comp_event_callback(cq);
-		spin_lock_irqsave(&cct->task_lock, flags_cct);
+		if (atomic_dec_and_test(&cq->nr_events))
+			wake_up(&cq->wait_completion);
 
-		spin_lock_irqsave(&cq->task_lock, flags_cq);
+		spin_lock_irq(&cct->task_lock);
+		spin_lock(&cq->task_lock);
 		cq->nr_callbacks--;
-		if (cq->nr_callbacks == 0) {
+		if (!cq->nr_callbacks) {
 			list_del_init(cct->cq_list.next);
 			cct->cq_jobs--;
 		}
-		spin_unlock_irqrestore(&cq->task_lock, flags_cq);
-
+		spin_unlock(&cq->task_lock);
 	}
-
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
-
-	return;
 }
 
-static int comp_task(void *__cct)
+static void comp_task_park(unsigned int cpu)
 {
-	struct ehca_cpu_comp_task* cct = __cct;
-	DECLARE_WAITQUEUE(wait, current);
-
-	set_current_state(TASK_INTERRUPTIBLE);
-	while(!kthread_should_stop()) {
-		add_wait_queue(&cct->wait_queue, &wait);
-
-		if (list_empty(&cct->cq_list))
-			schedule();
-		else
-			__set_current_state(TASK_RUNNING);
-
-		remove_wait_queue(&cct->wait_queue, &wait);
-
-		if (!list_empty(&cct->cq_list))
-			run_comp_task(__cct);
+	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+	struct ehca_cpu_comp_task *target;
+	struct task_struct *thread;
+	struct ehca_cq *cq, *tmp;
+	LIST_HEAD(list);
 
-		set_current_state(TASK_INTERRUPTIBLE);
+	spin_lock_irq(&cct->task_lock);
+	cct->cq_jobs = 0;
+	cct->active = 0;
+	list_splice_init(&cct->cq_list, &list);
+	spin_unlock_irq(&cct->task_lock);
+
+	cpu = find_next_online_cpu(pool);
+	target = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+	thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu);
+	spin_lock_irq(&target->task_lock);
+	list_for_each_entry_safe(cq, tmp, &list, entry) {
+		list_del(&cq->entry);
+		__queue_comp_task(cq, target, thread);
 	}
-	__set_current_state(TASK_RUNNING);
-
-	return 0;
+	spin_unlock_irq(&target->task_lock);
 }
 
-static struct task_struct *create_comp_task(struct ehca_comp_pool *pool,
-					    int cpu)
+static void comp_task_stop(unsigned int cpu, bool online)
 {
-	struct ehca_cpu_comp_task *cct;
-
-	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-	spin_lock_init(&cct->task_lock);
-	INIT_LIST_HEAD(&cct->cq_list);
-	init_waitqueue_head(&cct->wait_queue);
-	cct->task = kthread_create(comp_task, cct, "ehca_comp/%d", cpu);
-
-	return cct->task;
-}
-
-static void destroy_comp_task(struct ehca_comp_pool *pool,
-			      int cpu)
-{
-	struct ehca_cpu_comp_task *cct;
-	struct task_struct *task;
-	unsigned long flags_cct;
-
-	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
+	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 
-	task = cct->task;
-	cct->task = NULL;
+	spin_lock_irq(&cct->task_lock);
 	cct->cq_jobs = 0;
-
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
-
-	if (task)
-		kthread_stop(task);
-
-	return;
+	cct->active = 0;
+	WARN_ON(!list_empty(&cct->cq_list));
+	spin_unlock_irq(&cct->task_lock);
 }
 
-static void take_over_work(struct ehca_comp_pool *pool,
-			   int cpu)
+static int comp_task_should_run(unsigned int cpu)
 {
 	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-	LIST_HEAD(list);
-	struct ehca_cq *cq;
-	unsigned long flags_cct;
-
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
-
-	list_splice_init(&cct->cq_list, &list);
-
-	while(!list_empty(&list)) {
-	       cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
-
-	       list_del(&cq->entry);
-	       __queue_comp_task(cq, per_cpu_ptr(pool->cpu_comp_tasks,
-						 smp_processor_id()));
-	}
-
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
 
+	return cct->cq_jobs;
 }
 
-static int comp_pool_callback(struct notifier_block *nfb,
-			      unsigned long action,
-			      void *hcpu)
+static void comp_task(unsigned int cpu)
 {
-	unsigned int cpu = (unsigned long)hcpu;
-	struct ehca_cpu_comp_task *cct;
-
-	switch (action) {
-	case CPU_UP_PREPARE:
-		ehca_gen_dbg("CPU: %x (CPU_PREPARE)", cpu);
-		if(!create_comp_task(pool, cpu)) {
-			ehca_gen_err("Can't create comp_task for cpu: %x", cpu);
-			return NOTIFY_BAD;
-		}
-		break;
-	case CPU_UP_CANCELED:
-		ehca_gen_dbg("CPU: %x (CPU_CANCELED)", cpu);
-		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-		kthread_bind(cct->task, any_online_cpu(cpu_online_map));
-		destroy_comp_task(pool, cpu);
-		break;
-	case CPU_ONLINE:
-		ehca_gen_dbg("CPU: %x (CPU_ONLINE)", cpu);
-		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-		kthread_bind(cct->task, cpu);
-		wake_up_process(cct->task);
-		break;
-	case CPU_DOWN_PREPARE:
-		ehca_gen_dbg("CPU: %x (CPU_DOWN_PREPARE)", cpu);
-		break;
-	case CPU_DOWN_FAILED:
-		ehca_gen_dbg("CPU: %x (CPU_DOWN_FAILED)", cpu);
-		break;
-	case CPU_DEAD:
-		ehca_gen_dbg("CPU: %x (CPU_DEAD)", cpu);
-		destroy_comp_task(pool, cpu);
-		take_over_work(pool, cpu);
-		break;
+	struct ehca_cpu_comp_task *cct = this_cpu_ptr(pool->cpu_comp_tasks);
+	int cql_empty;
+
+	spin_lock_irq(&cct->task_lock);
+	cql_empty = list_empty(&cct->cq_list);
+	if (!cql_empty) {
+		__set_current_state(TASK_RUNNING);
+		run_comp_task(cct);
 	}
-
-	return NOTIFY_OK;
+	spin_unlock_irq(&cct->task_lock);
 }
 
-#endif
+static struct smp_hotplug_thread comp_pool_threads = {
+	.thread_should_run	= comp_task_should_run,
+	.thread_fn		= comp_task,
+	.thread_comm		= "ehca_comp/%u",
+	.cleanup		= comp_task_stop,
+	.park			= comp_task_park,
+};
 
 int ehca_create_comp_pool(void)
 {
-#ifdef CONFIG_INFINIBAND_EHCA_SCALING
-	int cpu;
-	struct task_struct *task;
+	int cpu, ret = -ENOMEM;
+
+	if (!ehca_scaling_code)
+		return 0;
 
 	pool = kzalloc(sizeof(struct ehca_comp_pool), GFP_KERNEL);
 	if (pool == NULL)
 		return -ENOMEM;
 
 	spin_lock_init(&pool->last_cpu_lock);
-	pool->last_cpu = any_online_cpu(cpu_online_map);
+	pool->last_cpu = cpumask_any(cpu_online_mask);
 
 	pool->cpu_comp_tasks = alloc_percpu(struct ehca_cpu_comp_task);
-	if (pool->cpu_comp_tasks == NULL) {
-		kfree(pool);
-		return -EINVAL;
-	}
+	if (!pool->cpu_comp_tasks)
+		goto out_pool;
 
-	for_each_online_cpu(cpu) {
-		task = create_comp_task(pool, cpu);
-		if (task) {
-			kthread_bind(task, cpu);
-			wake_up_process(task);
-		}
+	pool->cpu_comp_threads = alloc_percpu(struct task_struct *);
+	if (!pool->cpu_comp_threads)
+		goto out_tasks;
+
+	for_each_present_cpu(cpu) {
+		struct ehca_cpu_comp_task *cct;
+
+		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+		spin_lock_init(&cct->task_lock);
+		INIT_LIST_HEAD(&cct->cq_list);
 	}
 
-	comp_pool_callback_nb.notifier_call = comp_pool_callback;
-	comp_pool_callback_nb.priority =0;
-	register_cpu_notifier(&comp_pool_callback_nb);
-#endif
+	comp_pool_threads.store = pool->cpu_comp_threads;
+	ret = smpboot_register_percpu_thread(&comp_pool_threads);
+	if (ret)
+		goto out_threads;
 
-	return 0;
+	pr_info("eHCA scaling code enabled\n");
+	return ret;
+
+out_threads:
+	free_percpu(pool->cpu_comp_threads);
+out_tasks:
+	free_percpu(pool->cpu_comp_tasks);
+out_pool:
+	kfree(pool);
+	return ret;
 }
 
 void ehca_destroy_comp_pool(void)
 {
-#ifdef CONFIG_INFINIBAND_EHCA_SCALING
-	int i;
-
-	unregister_cpu_notifier(&comp_pool_callback_nb);
+	if (!ehca_scaling_code)
+		return;
 
-	for (i = 0; i < NR_CPUS; i++) {
-		if (cpu_online(i))
-			destroy_comp_task(pool, i);
-	}
-#endif
+	smpboot_unregister_percpu_thread(&comp_pool_threads);
 
-	return;
+	free_percpu(pool->cpu_comp_threads);
+	free_percpu(pool->cpu_comp_tasks);
+	kfree(pool);
 }