1 files changed, 555 insertions, 163 deletions

diff --git a/drivers/infiniband/hw/hfi1/affinity.c b/drivers/infiniband/hw/hfi1/affinity.c
index e2cd2cd3b28a..ee7fedc67b86 100644
--- a/drivers/infiniband/hw/hfi1/affinity.c
+++ b/drivers/infiniband/hw/hfi1/affinity.c
@@ -1,53 +1,12 @@
+// SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
 /*
- * Copyright(c) 2015, 2016 Intel Corporation.
- *
- * This file is provided under a dual BSD/GPLv2 license.  When using or
- * redistributing this file, you may do so under either license.
- *
- * GPL LICENSE SUMMARY
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of version 2 of the GNU General Public License as
- * published by the Free Software Foundation.
- *
- * This program is distributed in the hope that it will be useful, but
- * WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
- * General Public License for more details.
- *
- * BSD LICENSE
- *
- * 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.
- *  - Neither the name of Intel Corporation nor the names of its
- *    contributors may be used to endorse or promote products derived
- *    from this software without specific prior written permission.
- *
- * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
- * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
- * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
- * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
- * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
- * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
- * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
- * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
- * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
- * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
- * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
- *
+ * Copyright(c) 2015 - 2020 Intel Corporation.
  */
+
 #include <linux/topology.h>
 #include <linux/cpumask.h>
-#include <linux/module.h>
 #include <linux/interrupt.h>
+#include <linux/numa.h>
 
 #include "hfi.h"
 #include "affinity.h"
@@ -63,6 +22,7 @@ struct hfi1_affinity_node_list node_affinity = {
 static const char * const irq_type_names[] = {
 	"SDMA",
 	"RCVCTXT",
+	"NETDEVCTXT",
 	"GENERAL",
 	"OTHER",
 };
@@ -77,12 +37,62 @@ static inline void init_cpu_mask_set(struct cpu_mask_set *set)
 	set->gen = 0;
 }
 
+/* Increment generation of CPU set if needed */
+static void _cpu_mask_set_gen_inc(struct cpu_mask_set *set)
+{
+	if (cpumask_equal(&set->mask, &set->used)) {
+		/*
+		 * We've used up all the CPUs, bump up the generation
+		 * and reset the 'used' map
+		 */
+		set->gen++;
+		cpumask_clear(&set->used);
+	}
+}
+
+static void _cpu_mask_set_gen_dec(struct cpu_mask_set *set)
+{
+	if (cpumask_empty(&set->used) && set->gen) {
+		set->gen--;
+		cpumask_copy(&set->used, &set->mask);
+	}
+}
+
+/* Get the first CPU from the list of unused CPUs in a CPU set data structure */
+static int cpu_mask_set_get_first(struct cpu_mask_set *set, cpumask_var_t diff)
+{
+	int cpu;
+
+	if (!diff || !set)
+		return -EINVAL;
+
+	_cpu_mask_set_gen_inc(set);
+
+	/* Find out CPUs left in CPU mask */
+	cpumask_andnot(diff, &set->mask, &set->used);
+
+	cpu = cpumask_first(diff);
+	if (cpu >= nr_cpu_ids) /* empty */
+		cpu = -EINVAL;
+	else
+		cpumask_set_cpu(cpu, &set->used);
+
+	return cpu;
+}
+
+static void cpu_mask_set_put(struct cpu_mask_set *set, int cpu)
+{
+	if (!set)
+		return;
+
+	cpumask_clear_cpu(cpu, &set->used);
+	_cpu_mask_set_gen_dec(set);
+}
+
 /* Initialize non-HT cpu cores mask */
 void init_real_cpu_mask(void)
 {
-	int possible, curr_cpu, i, ht;
-
-	cpumask_clear(&node_affinity.real_cpu_mask);
+	int possible, curr_cpu, ht;
 
 	/* Start with cpu online mask as the real cpu mask */
 	cpumask_copy(&node_affinity.real_cpu_mask, cpu_online_mask);
@@ -98,17 +108,10 @@ void init_real_cpu_mask(void)
 	 * "real" cores.  Assumes that HT cores are not enumerated in
 	 * succession (except in the single core case).
 	 */
-	curr_cpu = cpumask_first(&node_affinity.real_cpu_mask);
-	for (i = 0; i < possible / ht; i++)
-		curr_cpu = cpumask_next(curr_cpu, &node_affinity.real_cpu_mask);
-	/*
-	 * Step 2.  Remove the remaining HT siblings.  Use cpumask_next() to
-	 * skip any gaps.
-	 */
-	for (; i < possible; i++) {
-		cpumask_clear_cpu(curr_cpu, &node_affinity.real_cpu_mask);
-		curr_cpu = cpumask_next(curr_cpu, &node_affinity.real_cpu_mask);
-	}
+	curr_cpu = cpumask_nth(possible / ht, &node_affinity.real_cpu_mask) + 1;
+
+	/* Step 2.  Remove the remaining HT siblings. */
+	cpumask_clear_cpus(&node_affinity.real_cpu_mask, curr_cpu, nr_cpu_ids - curr_cpu);
 }
 
 int node_affinity_init(void)
@@ -146,7 +149,7 @@ int node_affinity_init(void)
 		while ((dev = pci_get_device(ids->vendor, ids->device, dev))) {
 			node = pcibus_to_node(dev->bus);
 			if (node < 0)
-				node = numa_node_id();
+				goto out;
 
 			hfi1_per_node_cntr[node]++;
 		}
@@ -154,9 +157,29 @@ int node_affinity_init(void)
 	}
 
 	return 0;
+
+out:
+	/*
+	 * Invalid PCI NUMA node information found, note it, and populate
+	 * our database 1:1.
+	 */
+	pr_err("HFI: Invalid PCI NUMA node. Performance may be affected\n");
+	pr_err("HFI: System BIOS may need to be upgraded\n");
+	for (node = 0; node < node_affinity.num_possible_nodes; node++)
+		hfi1_per_node_cntr[node] = 1;
+
+	pci_dev_put(dev);
+
+	return 0;
 }
 
-void node_affinity_destroy(void)
+static void node_affinity_destroy(struct hfi1_affinity_node *entry)
+{
+	free_percpu(entry->comp_vect_affinity);
+	kfree(entry);
+}
+
+void node_affinity_destroy_all(void)
 {
 	struct list_head *pos, *q;
 	struct hfi1_affinity_node *entry;
@@ -166,7 +189,7 @@ void node_affinity_destroy(void)
 		entry = list_entry(pos, struct hfi1_affinity_node,
 				   list);
 		list_del(pos);
-		kfree(entry);
+		node_affinity_destroy(entry);
 	}
 	mutex_unlock(&node_affinity.lock);
 	kfree(hfi1_per_node_cntr);
@@ -180,6 +203,7 @@ static struct hfi1_affinity_node *node_affinity_allocate(int node)
 	if (!entry)
 		return NULL;
 	entry->node = node;
+	entry->comp_vect_affinity = alloc_percpu(u16);
 	INIT_LIST_HEAD(&entry->list);
 
 	return entry;
@@ -197,11 +221,9 @@ static void node_affinity_add_tail(struct hfi1_affinity_node *entry)
 /* It must be called with node_affinity.lock held */
 static struct hfi1_affinity_node *node_affinity_lookup(int node)
 {
-	struct list_head *pos;
 	struct hfi1_affinity_node *entry;
 
-	list_for_each(pos, &node_affinity.list) {
-		entry = list_entry(pos, struct hfi1_affinity_node, list);
+	list_for_each_entry(entry, &node_affinity.list, list) {
 		if (entry->node == node)
 			return entry;
 	}
@@ -209,6 +231,344 @@ static struct hfi1_affinity_node *node_affinity_lookup(int node)
 	return NULL;
 }
 
+static int per_cpu_affinity_get(cpumask_var_t possible_cpumask,
+				u16 __percpu *comp_vect_affinity)
+{
+	int curr_cpu;
+	u16 cntr;
+	u16 prev_cntr;
+	int ret_cpu;
+
+	if (!possible_cpumask) {
+		ret_cpu = -EINVAL;
+		goto fail;
+	}
+
+	if (!comp_vect_affinity) {
+		ret_cpu = -EINVAL;
+		goto fail;
+	}
+
+	ret_cpu = cpumask_first(possible_cpumask);
+	if (ret_cpu >= nr_cpu_ids) {
+		ret_cpu = -EINVAL;
+		goto fail;
+	}
+
+	prev_cntr = *per_cpu_ptr(comp_vect_affinity, ret_cpu);
+	for_each_cpu(curr_cpu, possible_cpumask) {
+		cntr = *per_cpu_ptr(comp_vect_affinity, curr_cpu);
+
+		if (cntr < prev_cntr) {
+			ret_cpu = curr_cpu;
+			prev_cntr = cntr;
+		}
+	}
+
+	*per_cpu_ptr(comp_vect_affinity, ret_cpu) += 1;
+
+fail:
+	return ret_cpu;
+}
+
+static int per_cpu_affinity_put_max(cpumask_var_t possible_cpumask,
+				    u16 __percpu *comp_vect_affinity)
+{
+	int curr_cpu;
+	int max_cpu;
+	u16 cntr;
+	u16 prev_cntr;
+
+	if (!possible_cpumask)
+		return -EINVAL;
+
+	if (!comp_vect_affinity)
+		return -EINVAL;
+
+	max_cpu = cpumask_first(possible_cpumask);
+	if (max_cpu >= nr_cpu_ids)
+		return -EINVAL;
+
+	prev_cntr = *per_cpu_ptr(comp_vect_affinity, max_cpu);
+	for_each_cpu(curr_cpu, possible_cpumask) {
+		cntr = *per_cpu_ptr(comp_vect_affinity, curr_cpu);
+
+		if (cntr > prev_cntr) {
+			max_cpu = curr_cpu;
+			prev_cntr = cntr;
+		}
+	}
+
+	*per_cpu_ptr(comp_vect_affinity, max_cpu) -= 1;
+
+	return max_cpu;
+}
+
+/*
+ * Non-interrupt CPUs are used first, then interrupt CPUs.
+ * Two already allocated cpu masks must be passed.
+ */
+static int _dev_comp_vect_cpu_get(struct hfi1_devdata *dd,
+				  struct hfi1_affinity_node *entry,
+				  cpumask_var_t non_intr_cpus,
+				  cpumask_var_t available_cpus)
+	__must_hold(&node_affinity.lock)
+{
+	int cpu;
+	struct cpu_mask_set *set = dd->comp_vect;
+
+	lockdep_assert_held(&node_affinity.lock);
+	if (!non_intr_cpus) {
+		cpu = -1;
+		goto fail;
+	}
+
+	if (!available_cpus) {
+		cpu = -1;
+		goto fail;
+	}
+
+	/* Available CPUs for pinning completion vectors */
+	_cpu_mask_set_gen_inc(set);
+	cpumask_andnot(available_cpus, &set->mask, &set->used);
+
+	/* Available CPUs without SDMA engine interrupts */
+	cpumask_andnot(non_intr_cpus, available_cpus,
+		       &entry->def_intr.used);
+
+	/* If there are non-interrupt CPUs available, use them first */
+	cpu = cpumask_first(non_intr_cpus);
+
+	/* Otherwise, use interrupt CPUs */
+	if (cpu >= nr_cpu_ids)
+		cpu = cpumask_first(available_cpus);
+
+	if (cpu >= nr_cpu_ids) { /* empty */
+		cpu = -1;
+		goto fail;
+	}
+	cpumask_set_cpu(cpu, &set->used);
+
+fail:
+	return cpu;
+}
+
+static void _dev_comp_vect_cpu_put(struct hfi1_devdata *dd, int cpu)
+{
+	struct cpu_mask_set *set = dd->comp_vect;
+
+	if (cpu < 0)
+		return;
+
+	cpu_mask_set_put(set, cpu);
+}
+
+/* _dev_comp_vect_mappings_destroy() is reentrant */
+static void _dev_comp_vect_mappings_destroy(struct hfi1_devdata *dd)
+{
+	int i, cpu;
+
+	if (!dd->comp_vect_mappings)
+		return;
+
+	for (i = 0; i < dd->comp_vect_possible_cpus; i++) {
+		cpu = dd->comp_vect_mappings[i];
+		_dev_comp_vect_cpu_put(dd, cpu);
+		dd->comp_vect_mappings[i] = -1;
+		hfi1_cdbg(AFFINITY,
+			  "[%s] Release CPU %d from completion vector %d",
+			  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), cpu, i);
+	}
+
+	kfree(dd->comp_vect_mappings);
+	dd->comp_vect_mappings = NULL;
+}
+
+/*
+ * This function creates the table for looking up CPUs for completion vectors.
+ * num_comp_vectors needs to have been initilized before calling this function.
+ */
+static int _dev_comp_vect_mappings_create(struct hfi1_devdata *dd,
+					  struct hfi1_affinity_node *entry)
+	__must_hold(&node_affinity.lock)
+{
+	int i, cpu, ret;
+	cpumask_var_t non_intr_cpus;
+	cpumask_var_t available_cpus;
+
+	lockdep_assert_held(&node_affinity.lock);
+
+	if (!zalloc_cpumask_var(&non_intr_cpus, GFP_KERNEL))
+		return -ENOMEM;
+
+	if (!zalloc_cpumask_var(&available_cpus, GFP_KERNEL)) {
+		free_cpumask_var(non_intr_cpus);
+		return -ENOMEM;
+	}
+
+	dd->comp_vect_mappings = kcalloc(dd->comp_vect_possible_cpus,
+					 sizeof(*dd->comp_vect_mappings),
+					 GFP_KERNEL);
+	if (!dd->comp_vect_mappings) {
+		ret = -ENOMEM;
+		goto fail;
+	}
+	for (i = 0; i < dd->comp_vect_possible_cpus; i++)
+		dd->comp_vect_mappings[i] = -1;
+
+	for (i = 0; i < dd->comp_vect_possible_cpus; i++) {
+		cpu = _dev_comp_vect_cpu_get(dd, entry, non_intr_cpus,
+					     available_cpus);
+		if (cpu < 0) {
+			ret = -EINVAL;
+			goto fail;
+		}
+
+		dd->comp_vect_mappings[i] = cpu;
+		hfi1_cdbg(AFFINITY,
+			  "[%s] Completion Vector %d -> CPU %d",
+			  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), i, cpu);
+	}
+
+	free_cpumask_var(available_cpus);
+	free_cpumask_var(non_intr_cpus);
+	return 0;
+
+fail:
+	free_cpumask_var(available_cpus);
+	free_cpumask_var(non_intr_cpus);
+	_dev_comp_vect_mappings_destroy(dd);
+
+	return ret;
+}
+
+int hfi1_comp_vectors_set_up(struct hfi1_devdata *dd)
+{
+	int ret;
+	struct hfi1_affinity_node *entry;
+
+	mutex_lock(&node_affinity.lock);
+	entry = node_affinity_lookup(dd->node);
+	if (!entry) {
+		ret = -EINVAL;
+		goto unlock;
+	}
+	ret = _dev_comp_vect_mappings_create(dd, entry);
+unlock:
+	mutex_unlock(&node_affinity.lock);
+
+	return ret;
+}
+
+void hfi1_comp_vectors_clean_up(struct hfi1_devdata *dd)
+{
+	_dev_comp_vect_mappings_destroy(dd);
+}
+
+int hfi1_comp_vect_mappings_lookup(struct rvt_dev_info *rdi, int comp_vect)
+{
+	struct hfi1_ibdev *verbs_dev = dev_from_rdi(rdi);
+	struct hfi1_devdata *dd = dd_from_dev(verbs_dev);
+
+	if (!dd->comp_vect_mappings)
+		return -EINVAL;
+	if (comp_vect >= dd->comp_vect_possible_cpus)
+		return -EINVAL;
+
+	return dd->comp_vect_mappings[comp_vect];
+}
+
+/*
+ * It assumes dd->comp_vect_possible_cpus is available.
+ */
+static int _dev_comp_vect_cpu_mask_init(struct hfi1_devdata *dd,
+					struct hfi1_affinity_node *entry,
+					bool first_dev_init)
+	__must_hold(&node_affinity.lock)
+{
+	int i, j, curr_cpu;
+	int possible_cpus_comp_vect = 0;
+	struct cpumask *dev_comp_vect_mask = &dd->comp_vect->mask;
+
+	lockdep_assert_held(&node_affinity.lock);
+	/*
+	 * If there's only one CPU available for completion vectors, then
+	 * there will only be one completion vector available. Othewise,
+	 * the number of completion vector available will be the number of
+	 * available CPUs divide it by the number of devices in the
+	 * local NUMA node.
+	 */
+	if (cpumask_weight(&entry->comp_vect_mask) == 1) {
+		possible_cpus_comp_vect = 1;
+		dd_dev_warn(dd,
+			    "Number of kernel receive queues is too large for completion vector affinity to be effective\n");
+	} else {
+		possible_cpus_comp_vect +=
+			cpumask_weight(&entry->comp_vect_mask) /
+				       hfi1_per_node_cntr[dd->node];
+
+		/*
+		 * If the completion vector CPUs available doesn't divide
+		 * evenly among devices, then the first device device to be
+		 * initialized gets an extra CPU.
+		 */
+		if (first_dev_init &&
+		    cpumask_weight(&entry->comp_vect_mask) %
+		    hfi1_per_node_cntr[dd->node] != 0)
+			possible_cpus_comp_vect++;
+	}
+
+	dd->comp_vect_possible_cpus = possible_cpus_comp_vect;
+
+	/* Reserving CPUs for device completion vector */
+	for (i = 0; i < dd->comp_vect_possible_cpus; i++) {
+		curr_cpu = per_cpu_affinity_get(&entry->comp_vect_mask,
+						entry->comp_vect_affinity);
+		if (curr_cpu < 0)
+			goto fail;
+
+		cpumask_set_cpu(curr_cpu, dev_comp_vect_mask);
+	}
+
+	hfi1_cdbg(AFFINITY,
+		  "[%s] Completion vector affinity CPU set(s) %*pbl",
+		  rvt_get_ibdev_name(&(dd)->verbs_dev.rdi),
+		  cpumask_pr_args(dev_comp_vect_mask));
+
+	return 0;
+
+fail:
+	for (j = 0; j < i; j++)
+		per_cpu_affinity_put_max(&entry->comp_vect_mask,
+					 entry->comp_vect_affinity);
+
+	return curr_cpu;
+}
+
+/*
+ * It assumes dd->comp_vect_possible_cpus is available.
+ */
+static void _dev_comp_vect_cpu_mask_clean_up(struct hfi1_devdata *dd,
+					     struct hfi1_affinity_node *entry)
+	__must_hold(&node_affinity.lock)
+{
+	int i, cpu;
+
+	lockdep_assert_held(&node_affinity.lock);
+	if (!dd->comp_vect_possible_cpus)
+		return;
+
+	for (i = 0; i < dd->comp_vect_possible_cpus; i++) {
+		cpu = per_cpu_affinity_put_max(&dd->comp_vect->mask,
+					       entry->comp_vect_affinity);
+		/* Clearing CPU in device completion vector cpu mask */
+		if (cpu >= 0)
+			cpumask_clear_cpu(cpu, &dd->comp_vect->mask);
+	}
+
+	dd->comp_vect_possible_cpus = 0;
+}
+
 /*
  * Interrupt affinity.
  *
@@ -222,14 +582,10 @@ static struct hfi1_affinity_node *node_affinity_lookup(int node)
  */
 int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 {
-	int node = pcibus_to_node(dd->pcidev->bus);
 	struct hfi1_affinity_node *entry;
 	const struct cpumask *local_mask;
-	int curr_cpu, possible, i;
-
-	if (node < 0)
-		node = numa_node_id();
-	dd->node = node;
+	int curr_cpu, possible, i, ret;
+	bool new_entry = false;
 
 	local_mask = cpumask_of_node(dd->node);
 	if (cpumask_first(local_mask) >= nr_cpu_ids)
@@ -243,15 +599,18 @@ int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 	 * create an entry in the global affinity structure and initialize it.
 	 */
 	if (!entry) {
-		entry = node_affinity_allocate(node);
+		entry = node_affinity_allocate(dd->node);
 		if (!entry) {
 			dd_dev_err(dd,
 				   "Unable to allocate global affinity node\n");
-			mutex_unlock(&node_affinity.lock);
-			return -ENOMEM;
+			ret = -ENOMEM;
+			goto fail;
 		}
+		new_entry = true;
+
 		init_cpu_mask_set(&entry->def_intr);
 		init_cpu_mask_set(&entry->rcv_intr);
+		cpumask_clear(&entry->comp_vect_mask);
 		cpumask_clear(&entry->general_intr_mask);
 		/* Use the "real" cpu mask of this node as the default */
 		cpumask_and(&entry->def_intr.mask, &node_affinity.real_cpu_mask,
@@ -299,15 +658,69 @@ int hfi1_dev_affinity_init(struct hfi1_devdata *dd)
 			 * engines, use the same CPU cores as general/control
 			 * context.
 			 */
-			if (cpumask_weight(&entry->def_intr.mask) == 0)
+			if (cpumask_empty(&entry->def_intr.mask))
 				cpumask_copy(&entry->def_intr.mask,
 					     &entry->general_intr_mask);
 		}
 
-		node_affinity_add_tail(entry);
+		/* Determine completion vector CPUs for the entire node */
+		cpumask_and(&entry->comp_vect_mask,
+			    &node_affinity.real_cpu_mask, local_mask);
+		cpumask_andnot(&entry->comp_vect_mask,
+			       &entry->comp_vect_mask,
+			       &entry->rcv_intr.mask);
+		cpumask_andnot(&entry->comp_vect_mask,
+			       &entry->comp_vect_mask,
+			       &entry->general_intr_mask);
+
+		/*
+		 * If there ends up being 0 CPU cores leftover for completion
+		 * vectors, use the same CPU core as the general/control
+		 * context.
+		 */
+		if (cpumask_empty(&entry->comp_vect_mask))
+			cpumask_copy(&entry->comp_vect_mask,
+				     &entry->general_intr_mask);
 	}
+
+	ret = _dev_comp_vect_cpu_mask_init(dd, entry, new_entry);
+	if (ret < 0)
+		goto fail;
+
+	if (new_entry)
+		node_affinity_add_tail(entry);
+
+	dd->affinity_entry = entry;
 	mutex_unlock(&node_affinity.lock);
+
 	return 0;
+
+fail:
+	if (new_entry)
+		node_affinity_destroy(entry);
+	mutex_unlock(&node_affinity.lock);
+	return ret;
+}
+
+void hfi1_dev_affinity_clean_up(struct hfi1_devdata *dd)
+{
+	struct hfi1_affinity_node *entry;
+
+	mutex_lock(&node_affinity.lock);
+	if (!dd->affinity_entry)
+		goto unlock;
+	entry = node_affinity_lookup(dd->node);
+	if (!entry)
+		goto unlock;
+
+	/*
+	 * Free device completion vector CPUs to be used by future
+	 * completion vectors
+	 */
+	_dev_comp_vect_cpu_mask_clean_up(dd, entry);
+unlock:
+	dd->affinity_entry = NULL;
+	mutex_unlock(&node_affinity.lock);
 }
 
 /*
@@ -335,10 +748,10 @@ static void hfi1_update_sdma_affinity(struct hfi1_msix_entry *msix, int cpu)
 	sde->cpu = cpu;
 	cpumask_clear(&msix->mask);
 	cpumask_set_cpu(cpu, &msix->mask);
-	dd_dev_dbg(dd, "IRQ vector: %u, type %s engine %u -> cpu: %d\n",
-		   msix->msix.vector, irq_type_names[msix->type],
+	dd_dev_dbg(dd, "IRQ: %u, type %s engine %u -> cpu: %d\n",
+		   msix->irq, irq_type_names[msix->type],
 		   sde->this_idx, cpu);
-	irq_set_affinity_hint(msix->msix.vector, &msix->mask);
+	irq_set_affinity_hint(msix->irq, &msix->mask);
 
 	/*
 	 * Set the new cpu in the hfi1_affinity_node and clean
@@ -347,10 +760,10 @@ static void hfi1_update_sdma_affinity(struct hfi1_msix_entry *msix, int cpu)
 	set = &entry->def_intr;
 	cpumask_set_cpu(cpu, &set->mask);
 	cpumask_set_cpu(cpu, &set->used);
-	for (i = 0; i < dd->num_msix_entries; i++) {
+	for (i = 0; i < dd->msix_info.max_requested; i++) {
 		struct hfi1_msix_entry *other_msix;
 
-		other_msix = &dd->msix_entries[i];
+		other_msix = &dd->msix_info.msix_entries[i];
 		if (other_msix->type != IRQ_SDMA || other_msix == msix)
 			continue;
 
@@ -387,7 +800,7 @@ static void hfi1_setup_sdma_notifier(struct hfi1_msix_entry *msix)
 {
 	struct irq_affinity_notify *notify = &msix->notify;
 
-	notify->irq = msix->msix.vector;
+	notify->irq = msix->irq;
 	notify->notify = hfi1_irq_notifier_notify;
 	notify->release = hfi1_irq_notifier_release;
 
@@ -412,7 +825,6 @@ static void hfi1_cleanup_sdma_notifier(struct hfi1_msix_entry *msix)
 static int get_irq_affinity(struct hfi1_devdata *dd,
 			    struct hfi1_msix_entry *msix)
 {
-	int ret;
 	cpumask_var_t diff;
 	struct hfi1_affinity_node *entry;
 	struct cpu_mask_set *set = NULL;
@@ -424,10 +836,6 @@ static int get_irq_affinity(struct hfi1_devdata *dd,
 	extra[0] = '\0';
 	cpumask_clear(&msix->mask);
 
-	ret = zalloc_cpumask_var(&diff, GFP_KERNEL);
-	if (!ret)
-		return -ENOMEM;
-
 	entry = node_affinity_lookup(dd->node);
 
 	switch (msix->type) {
@@ -447,6 +855,11 @@ static int get_irq_affinity(struct hfi1_devdata *dd,
 			set = &entry->rcv_intr;
 		scnprintf(extra, 64, "ctxt %u", rcd->ctxt);
 		break;
+	case IRQ_NETDEVCTXT:
+		rcd = (struct hfi1_ctxtdata *)msix->arg;
+		set = &entry->def_intr;
+		scnprintf(extra, 64, "ctxt %u", rcd->ctxt);
+		break;
 	default:
 		dd_dev_err(dd, "Invalid IRQ type %d\n", msix->type);
 		return -EINVAL;
@@ -458,31 +871,30 @@ static int get_irq_affinity(struct hfi1_devdata *dd,
 	 * finds its CPU here.
 	 */
 	if (cpu == -1 && set) {
-		if (cpumask_equal(&set->mask, &set->used)) {
-			/*
-			 * We've used up all the CPUs, bump up the generation
-			 * and reset the 'used' map
-			 */
-			set->gen++;
-			cpumask_clear(&set->used);
+		if (!zalloc_cpumask_var(&diff, GFP_KERNEL))
+			return -ENOMEM;
+
+		cpu = cpu_mask_set_get_first(set, diff);
+		if (cpu < 0) {
+			free_cpumask_var(diff);
+			dd_dev_err(dd, "Failure to obtain CPU for IRQ\n");
+			return cpu;
 		}
-		cpumask_andnot(diff, &set->mask, &set->used);
-		cpu = cpumask_first(diff);
-		cpumask_set_cpu(cpu, &set->used);
+
+		free_cpumask_var(diff);
 	}
 
 	cpumask_set_cpu(cpu, &msix->mask);
-	dd_dev_info(dd, "IRQ vector: %u, type %s %s -> cpu: %d\n",
-		    msix->msix.vector, irq_type_names[msix->type],
+	dd_dev_info(dd, "IRQ: %u, type %s %s -> cpu: %d\n",
+		    msix->irq, irq_type_names[msix->type],
 		    extra, cpu);
-	irq_set_affinity_hint(msix->msix.vector, &msix->mask);
+	irq_set_affinity_hint(msix->irq, &msix->mask);
 
 	if (msix->type == IRQ_SDMA) {
 		sde->cpu = cpu;
 		hfi1_setup_sdma_notifier(msix);
 	}
 
-	free_cpumask_var(diff);
 	return 0;
 }
 
@@ -500,7 +912,6 @@ void hfi1_put_irq_affinity(struct hfi1_devdata *dd,
 			   struct hfi1_msix_entry *msix)
 {
 	struct cpu_mask_set *set = NULL;
-	struct hfi1_ctxtdata *rcd;
 	struct hfi1_affinity_node *entry;
 
 	mutex_lock(&node_affinity.lock);
@@ -514,12 +925,17 @@ void hfi1_put_irq_affinity(struct hfi1_devdata *dd,
 	case IRQ_GENERAL:
 		/* Don't do accounting for general contexts */
 		break;
-	case IRQ_RCVCTXT:
-		rcd = (struct hfi1_ctxtdata *)msix->arg;
+	case IRQ_RCVCTXT: {
+		struct hfi1_ctxtdata *rcd = msix->arg;
+
 		/* Don't do accounting for control contexts */
 		if (rcd->ctxt != HFI1_CTRL_CTXT)
 			set = &entry->rcv_intr;
 		break;
+	}
+	case IRQ_NETDEVCTXT:
+		set = &entry->def_intr;
+		break;
 	default:
 		mutex_unlock(&node_affinity.lock);
 		return;
@@ -527,13 +943,10 @@ void hfi1_put_irq_affinity(struct hfi1_devdata *dd,
 
 	if (set) {
 		cpumask_andnot(&set->used, &set->used, &msix->mask);
-		if (cpumask_empty(&set->used) && set->gen) {
-			set->gen--;
-			cpumask_copy(&set->used, &set->mask);
-		}
+		_cpu_mask_set_gen_dec(set);
 	}
 
-	irq_set_affinity_hint(msix->msix.vector, NULL);
+	irq_set_affinity_hint(msix->irq, NULL);
 	cpumask_clear(&msix->mask);
 	mutex_unlock(&node_affinity.lock);
 }
@@ -542,32 +955,23 @@ void hfi1_put_irq_affinity(struct hfi1_devdata *dd,
 static void find_hw_thread_mask(uint hw_thread_no, cpumask_var_t hw_thread_mask,
 				struct hfi1_affinity_node_list *affinity)
 {
-	int possible, curr_cpu, i;
-	uint num_cores_per_socket = node_affinity.num_online_cpus /
-					affinity->num_core_siblings /
-						node_affinity.num_online_nodes;
+	int curr_cpu;
+	uint num_cores;
 
 	cpumask_copy(hw_thread_mask, &affinity->proc.mask);
-	if (affinity->num_core_siblings > 0) {
-		/* Removing other siblings not needed for now */
-		possible = cpumask_weight(hw_thread_mask);
-		curr_cpu = cpumask_first(hw_thread_mask);
-		for (i = 0;
-		     i < num_cores_per_socket * node_affinity.num_online_nodes;
-		     i++)
-			curr_cpu = cpumask_next(curr_cpu, hw_thread_mask);
-
-		for (; i < possible; i++) {
-			cpumask_clear_cpu(curr_cpu, hw_thread_mask);
-			curr_cpu = cpumask_next(curr_cpu, hw_thread_mask);
-		}
 
-		/* Identifying correct HW threads within physical cores */
-		cpumask_shift_left(hw_thread_mask, hw_thread_mask,
-				   num_cores_per_socket *
-				   node_affinity.num_online_nodes *
-				   hw_thread_no);
-	}
+	if (affinity->num_core_siblings == 0)
+		return;
+
+	num_cores = rounddown(node_affinity.num_online_cpus / affinity->num_core_siblings,
+				node_affinity.num_online_nodes);
+
+	/* Removing other siblings not needed for now */
+	curr_cpu = cpumask_nth(num_cores * node_affinity.num_online_nodes, hw_thread_mask) + 1;
+	cpumask_clear_cpus(hw_thread_mask, curr_cpu, nr_cpu_ids - curr_cpu);
+
+	/* Identifying correct HW threads within physical cores */
+	cpumask_shift_left(hw_thread_mask, hw_thread_mask, num_cores * hw_thread_no);
 }
 
 int hfi1_get_proc_affinity(int node)
@@ -576,7 +980,7 @@ int hfi1_get_proc_affinity(int node)
 	struct hfi1_affinity_node *entry;
 	cpumask_var_t diff, hw_thread_mask, available_mask, intrs_mask;
 	const struct cpumask *node_mask,
-		*proc_mask = &current->cpus_allowed;
+		*proc_mask = current->cpus_ptr;
 	struct hfi1_affinity_node_list *affinity = &node_affinity;
 	struct cpu_mask_set *set = &affinity->proc;
 
@@ -584,7 +988,7 @@ int hfi1_get_proc_affinity(int node)
 	 * check whether process/context affinity has already
 	 * been set
 	 */
-	if (cpumask_weight(proc_mask) == 1) {
+	if (current->nr_cpus_allowed == 1) {
 		hfi1_cdbg(PROC, "PID %u %s affinity set to CPU %*pbl",
 			  current->pid, current->comm,
 			  cpumask_pr_args(proc_mask));
@@ -595,7 +999,7 @@ int hfi1_get_proc_affinity(int node)
 		cpu = cpumask_first(proc_mask);
 		cpumask_set_cpu(cpu, &set->used);
 		goto done;
-	} else if (cpumask_weight(proc_mask) < cpumask_weight(&set->mask)) {
+	} else if (current->nr_cpus_allowed < cpumask_weight(&set->mask)) {
 		hfi1_cdbg(PROC, "PID %u %s affinity set to CPU set(s) %*pbl",
 			  current->pid, current->comm,
 			  cpumask_pr_args(proc_mask));
@@ -641,10 +1045,7 @@ int hfi1_get_proc_affinity(int node)
 	 * If we've used all available HW threads, clear the mask and start
 	 * overloading.
 	 */
-	if (cpumask_equal(&set->mask, &set->used)) {
-		set->gen++;
-		cpumask_clear(&set->used);
-	}
+	_cpu_mask_set_gen_inc(set);
 
 	/*
 	 * If NUMA node has CPUs used by interrupt handlers, include them in the
@@ -669,22 +1070,19 @@ int hfi1_get_proc_affinity(int node)
 	 * If HT cores are enabled, identify which HW threads within the
 	 * physical cores should be used.
 	 */
-	if (affinity->num_core_siblings > 0) {
-		for (i = 0; i < affinity->num_core_siblings; i++) {
-			find_hw_thread_mask(i, hw_thread_mask, affinity);
+	for (i = 0; i < affinity->num_core_siblings; i++) {
+		find_hw_thread_mask(i, hw_thread_mask, affinity);
 
-			/*
-			 * If there's at least one available core for this HW
-			 * thread number, stop looking for a core.
-			 *
-			 * diff will always be not empty at least once in this
-			 * loop as the used mask gets reset when
-			 * (set->mask == set->used) before this loop.
-			 */
-			cpumask_andnot(diff, hw_thread_mask, &set->used);
-			if (!cpumask_empty(diff))
-				break;
-		}
+		/*
+		 * If there's at least one available core for this HW
+		 * thread number, stop looking for a core.
+		 *
+		 * diff will always be not empty at least once in this
+		 * loop as the used mask gets reset when
+		 * (set->mask == set->used) before this loop.
+		 */
+		if (cpumask_andnot(diff, hw_thread_mask, &set->used))
+			break;
 	}
 	hfi1_cdbg(PROC, "Same available HW thread on all physical CPUs: %*pbl",
 		  cpumask_pr_args(hw_thread_mask));
@@ -715,8 +1113,7 @@ int hfi1_get_proc_affinity(int node)
 	 *    used for process assignments using the same method as
 	 *    the preferred NUMA node.
 	 */
-	cpumask_andnot(diff, available_mask, intrs_mask);
-	if (!cpumask_empty(diff))
+	if (cpumask_andnot(diff, available_mask, intrs_mask))
 		cpumask_copy(available_mask, diff);
 
 	/* If we don't have CPUs on the preferred node, use other NUMA nodes */
@@ -732,8 +1129,7 @@ int hfi1_get_proc_affinity(int node)
 		 * At first, we don't want to place processes on the same
 		 * CPUs as interrupt handlers.
 		 */
-		cpumask_andnot(diff, available_mask, intrs_mask);
-		if (!cpumask_empty(diff))
+		if (cpumask_andnot(diff, available_mask, intrs_mask))
 			cpumask_copy(available_mask, diff);
 	}
 	hfi1_cdbg(PROC, "Possible CPUs for process: %*pbl",
@@ -768,11 +1164,7 @@ void hfi1_put_proc_affinity(int cpu)
 		return;
 
 	mutex_lock(&affinity->lock);
-	cpumask_clear_cpu(cpu, &set->used);
+	cpu_mask_set_put(set, cpu);
 	hfi1_cdbg(PROC, "Returning CPU %d for future process assignment", cpu);
-	if (cpumask_empty(&set->used) && set->gen) {
-		set->gen--;
-		cpumask_copy(&set->used, &set->mask);
-	}
 	mutex_unlock(&affinity->lock);
 }