padata: Generic parallelization/serialization interface
Steffen Klassert [Wed, 6 Jan 2010 08:47:10 +0000 (19:47 +1100)]
This patch introduces an interface to process data objects
in parallel. The parallelized objects return after serialization
in the same order as they were before the parallelization.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

include/linux/padata.h [new file with mode: 0644]
init/Kconfig
kernel/Makefile
kernel/padata.c [new file with mode: 0644]

diff --git a/include/linux/padata.h b/include/linux/padata.h
new file mode 100644 (file)
index 0000000..51611da
--- /dev/null
@@ -0,0 +1,88 @@
+/*
+ * padata.h - header for the padata parallelization interface
+ *
+ * Copyright (C) 2008, 2009 secunet Security Networks AG
+ * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#ifndef PADATA_H
+#define PADATA_H
+
+#include <linux/workqueue.h>
+#include <linux/spinlock.h>
+#include <linux/list.h>
+
+struct padata_priv {
+       struct list_head        list;
+       struct parallel_data    *pd;
+       int                     cb_cpu;
+       int                     seq_nr;
+       int                     info;
+       void                    (*parallel)(struct padata_priv *padata);
+       void                    (*serial)(struct padata_priv *padata);
+};
+
+struct padata_list {
+       struct list_head        list;
+       spinlock_t              lock;
+};
+
+struct padata_queue {
+       struct padata_list      parallel;
+       struct padata_list      reorder;
+       struct padata_list      serial;
+       struct work_struct      pwork;
+       struct work_struct      swork;
+       struct parallel_data    *pd;
+       atomic_t                num_obj;
+       int                     cpu_index;
+};
+
+struct parallel_data {
+       struct padata_instance  *pinst;
+       struct padata_queue     *queue;
+       atomic_t                seq_nr;
+       atomic_t                reorder_objects;
+       atomic_t                refcnt;
+       unsigned int            max_seq_nr;
+       cpumask_var_t           cpumask;
+       spinlock_t              lock;
+};
+
+struct padata_instance {
+       struct notifier_block   cpu_notifier;
+       struct workqueue_struct *wq;
+       struct parallel_data    *pd;
+       cpumask_var_t           cpumask;
+       struct mutex            lock;
+       u8                      flags;
+#define        PADATA_INIT             1
+#define        PADATA_RESET            2
+};
+
+extern struct padata_instance *padata_alloc(const struct cpumask *cpumask,
+                                           struct workqueue_struct *wq);
+extern void padata_free(struct padata_instance *pinst);
+extern int padata_do_parallel(struct padata_instance *pinst,
+                             struct padata_priv *padata, int cb_cpu);
+extern void padata_do_serial(struct padata_priv *padata);
+extern int padata_set_cpumask(struct padata_instance *pinst,
+                             cpumask_var_t cpumask);
+extern int padata_add_cpu(struct padata_instance *pinst, int cpu);
+extern int padata_remove_cpu(struct padata_instance *pinst, int cpu);
+extern void padata_start(struct padata_instance *pinst);
+extern void padata_stop(struct padata_instance *pinst);
+#endif
index a23da9f..9fd23bc 100644 (file)
@@ -1252,4 +1252,8 @@ source "block/Kconfig"
 config PREEMPT_NOTIFIERS
        bool
 
+config PADATA
+       depends on SMP
+       bool
+
 source "kernel/Kconfig.locks"
index 864ff75..6aebdeb 100644 (file)
@@ -100,6 +100,7 @@ obj-$(CONFIG_SLOW_WORK_DEBUG) += slow-work-debugfs.o
 obj-$(CONFIG_PERF_EVENTS) += perf_event.o
 obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
 obj-$(CONFIG_USER_RETURN_NOTIFIER) += user-return-notifier.o
+obj-$(CONFIG_PADATA) += padata.o
 
 ifneq ($(CONFIG_SCHED_OMIT_FRAME_POINTER),y)
 # According to Alan Modra <alan@linuxcare.com.au>, the -fno-omit-frame-pointer is
diff --git a/kernel/padata.c b/kernel/padata.c
new file mode 100644 (file)
index 0000000..6f9bcb8
--- /dev/null
@@ -0,0 +1,690 @@
+/*
+ * padata.c - generic interface to process data streams in parallel
+ *
+ * Copyright (C) 2008, 2009 secunet Security Networks AG
+ * Copyright (C) 2008, 2009 Steffen Klassert <steffen.klassert@secunet.com>
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms and conditions of the GNU General Public License,
+ * version 2, as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope 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.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program; if not, write to the Free Software Foundation, Inc.,
+ * 51 Franklin St - Fifth Floor, Boston, MA 02110-1301 USA.
+ */
+
+#include <linux/module.h>
+#include <linux/cpumask.h>
+#include <linux/err.h>
+#include <linux/cpu.h>
+#include <linux/padata.h>
+#include <linux/mutex.h>
+#include <linux/sched.h>
+#include <linux/rcupdate.h>
+
+#define MAX_SEQ_NR INT_MAX - NR_CPUS
+#define MAX_OBJ_NUM 10000 * NR_CPUS
+
+static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
+{
+       int cpu, target_cpu;
+
+       target_cpu = cpumask_first(pd->cpumask);
+       for (cpu = 0; cpu < cpu_index; cpu++)
+               target_cpu = cpumask_next(target_cpu, pd->cpumask);
+
+       return target_cpu;
+}
+
+static int padata_cpu_hash(struct padata_priv *padata)
+{
+       int cpu_index;
+       struct parallel_data *pd;
+
+       pd =  padata->pd;
+
+       /*
+        * Hash the sequence numbers to the cpus by taking
+        * seq_nr mod. number of cpus in use.
+        */
+       cpu_index =  padata->seq_nr % cpumask_weight(pd->cpumask);
+
+       return padata_index_to_cpu(pd, cpu_index);
+}
+
+static void padata_parallel_worker(struct work_struct *work)
+{
+       struct padata_queue *queue;
+       struct parallel_data *pd;
+       struct padata_instance *pinst;
+       LIST_HEAD(local_list);
+
+       local_bh_disable();
+       queue = container_of(work, struct padata_queue, pwork);
+       pd = queue->pd;
+       pinst = pd->pinst;
+
+       spin_lock(&queue->parallel.lock);
+       list_replace_init(&queue->parallel.list, &local_list);
+       spin_unlock(&queue->parallel.lock);
+
+       while (!list_empty(&local_list)) {
+               struct padata_priv *padata;
+
+               padata = list_entry(local_list.next,
+                                   struct padata_priv, list);
+
+               list_del_init(&padata->list);
+
+               padata->parallel(padata);
+       }
+
+       local_bh_enable();
+}
+
+/*
+ * padata_do_parallel - padata parallelization function
+ *
+ * @pinst: padata instance
+ * @padata: object to be parallelized
+ * @cb_cpu: cpu the serialization callback function will run on,
+ *          must be in the cpumask of padata.
+ *
+ * The parallelization callback function will run with BHs off.
+ * Note: Every object which is parallelized by padata_do_parallel
+ * must be seen by padata_do_serial.
+ */
+int padata_do_parallel(struct padata_instance *pinst,
+                      struct padata_priv *padata, int cb_cpu)
+{
+       int target_cpu, err;
+       struct padata_queue *queue;
+       struct parallel_data *pd;
+
+       rcu_read_lock_bh();
+
+       pd = rcu_dereference(pinst->pd);
+
+       err = 0;
+       if (!(pinst->flags & PADATA_INIT))
+               goto out;
+
+       err =  -EBUSY;
+       if ((pinst->flags & PADATA_RESET))
+               goto out;
+
+       if (atomic_read(&pd->refcnt) >= MAX_OBJ_NUM)
+               goto out;
+
+       err = -EINVAL;
+       if (!cpumask_test_cpu(cb_cpu, pd->cpumask))
+               goto out;
+
+       err = -EINPROGRESS;
+       atomic_inc(&pd->refcnt);
+       padata->pd = pd;
+       padata->cb_cpu = cb_cpu;
+
+       if (unlikely(atomic_read(&pd->seq_nr) == pd->max_seq_nr))
+               atomic_set(&pd->seq_nr, -1);
+
+       padata->seq_nr = atomic_inc_return(&pd->seq_nr);
+
+       target_cpu = padata_cpu_hash(padata);
+       queue = per_cpu_ptr(pd->queue, target_cpu);
+
+       spin_lock(&queue->parallel.lock);
+       list_add_tail(&padata->list, &queue->parallel.list);
+       spin_unlock(&queue->parallel.lock);
+
+       queue_work_on(target_cpu, pinst->wq, &queue->pwork);
+
+out:
+       rcu_read_unlock_bh();
+
+       return err;
+}
+EXPORT_SYMBOL(padata_do_parallel);
+
+static struct padata_priv *padata_get_next(struct parallel_data *pd)
+{
+       int cpu, num_cpus, empty, calc_seq_nr;
+       int seq_nr, next_nr, overrun, next_overrun;
+       struct padata_queue *queue, *next_queue;
+       struct padata_priv *padata;
+       struct padata_list *reorder;
+
+       empty = 0;
+       next_nr = -1;
+       next_overrun = 0;
+       next_queue = NULL;
+
+       num_cpus = cpumask_weight(pd->cpumask);
+
+       for_each_cpu(cpu, pd->cpumask) {
+               queue = per_cpu_ptr(pd->queue, cpu);
+               reorder = &queue->reorder;
+
+               /*
+                * Calculate the seq_nr of the object that should be
+                * next in this queue.
+                */
+               overrun = 0;
+               calc_seq_nr = (atomic_read(&queue->num_obj) * num_cpus)
+                              + queue->cpu_index;
+
+               if (unlikely(calc_seq_nr > pd->max_seq_nr)) {
+                       calc_seq_nr = calc_seq_nr - pd->max_seq_nr - 1;
+                       overrun = 1;
+               }
+
+               if (!list_empty(&reorder->list)) {
+                       padata = list_entry(reorder->list.next,
+                                           struct padata_priv, list);
+
+                       seq_nr  = padata->seq_nr;
+                       BUG_ON(calc_seq_nr != seq_nr);
+               } else {
+                       seq_nr = calc_seq_nr;
+                       empty++;
+               }
+
+               if (next_nr < 0 || seq_nr < next_nr
+                   || (next_overrun && !overrun)) {
+                       next_nr = seq_nr;
+                       next_overrun = overrun;
+                       next_queue = queue;
+               }
+       }
+
+       padata = NULL;
+
+       if (empty == num_cpus)
+               goto out;
+
+       reorder = &next_queue->reorder;
+
+       if (!list_empty(&reorder->list)) {
+               padata = list_entry(reorder->list.next,
+                                   struct padata_priv, list);
+
+               if (unlikely(next_overrun)) {
+                       for_each_cpu(cpu, pd->cpumask) {
+                               queue = per_cpu_ptr(pd->queue, cpu);
+                               atomic_set(&queue->num_obj, 0);
+                       }
+               }
+
+               spin_lock(&reorder->lock);
+               list_del_init(&padata->list);
+               atomic_dec(&pd->reorder_objects);
+               spin_unlock(&reorder->lock);
+
+               atomic_inc(&next_queue->num_obj);
+
+               goto out;
+       }
+
+       if (next_nr % num_cpus == next_queue->cpu_index) {
+               padata = ERR_PTR(-ENODATA);
+               goto out;
+       }
+
+       padata = ERR_PTR(-EINPROGRESS);
+out:
+       return padata;
+}
+
+static void padata_reorder(struct parallel_data *pd)
+{
+       struct padata_priv *padata;
+       struct padata_queue *queue;
+       struct padata_instance *pinst = pd->pinst;
+
+try_again:
+       if (!spin_trylock_bh(&pd->lock))
+               goto out;
+
+       while (1) {
+               padata = padata_get_next(pd);
+
+               if (!padata || PTR_ERR(padata) == -EINPROGRESS)
+                       break;
+
+               if (PTR_ERR(padata) == -ENODATA) {
+                       spin_unlock_bh(&pd->lock);
+                       goto out;
+               }
+
+               queue = per_cpu_ptr(pd->queue, padata->cb_cpu);
+
+               spin_lock(&queue->serial.lock);
+               list_add_tail(&padata->list, &queue->serial.list);
+               spin_unlock(&queue->serial.lock);
+
+               queue_work_on(padata->cb_cpu, pinst->wq, &queue->swork);
+       }
+
+       spin_unlock_bh(&pd->lock);
+
+       if (atomic_read(&pd->reorder_objects))
+               goto try_again;
+
+out:
+       return;
+}
+
+static void padata_serial_worker(struct work_struct *work)
+{
+       struct padata_queue *queue;
+       struct parallel_data *pd;
+       LIST_HEAD(local_list);
+
+       local_bh_disable();
+       queue = container_of(work, struct padata_queue, swork);
+       pd = queue->pd;
+
+       spin_lock(&queue->serial.lock);
+       list_replace_init(&queue->serial.list, &local_list);
+       spin_unlock(&queue->serial.lock);
+
+       while (!list_empty(&local_list)) {
+               struct padata_priv *padata;
+
+               padata = list_entry(local_list.next,
+                                   struct padata_priv, list);
+
+               list_del_init(&padata->list);
+
+               padata->serial(padata);
+               atomic_dec(&pd->refcnt);
+       }
+       local_bh_enable();
+}
+
+/*
+ * padata_do_serial - padata serialization function
+ *
+ * @padata: object to be serialized.
+ *
+ * padata_do_serial must be called for every parallelized object.
+ * The serialization callback function will run with BHs off.
+ */
+void padata_do_serial(struct padata_priv *padata)
+{
+       int cpu;
+       struct padata_queue *queue;
+       struct parallel_data *pd;
+
+       pd = padata->pd;
+
+       cpu = get_cpu();
+       queue = per_cpu_ptr(pd->queue, cpu);
+
+       spin_lock(&queue->reorder.lock);
+       atomic_inc(&pd->reorder_objects);
+       list_add_tail(&padata->list, &queue->reorder.list);
+       spin_unlock(&queue->reorder.lock);
+
+       put_cpu();
+
+       padata_reorder(pd);
+}
+EXPORT_SYMBOL(padata_do_serial);
+
+static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
+                                            const struct cpumask *cpumask)
+{
+       int cpu, cpu_index, num_cpus;
+       struct padata_queue *queue;
+       struct parallel_data *pd;
+
+       cpu_index = 0;
+
+       pd = kzalloc(sizeof(struct parallel_data), GFP_KERNEL);
+       if (!pd)
+               goto err;
+
+       pd->queue = alloc_percpu(struct padata_queue);
+       if (!pd->queue)
+               goto err_free_pd;
+
+       if (!alloc_cpumask_var(&pd->cpumask, GFP_KERNEL))
+               goto err_free_queue;
+
+       for_each_possible_cpu(cpu) {
+               queue = per_cpu_ptr(pd->queue, cpu);
+
+               queue->pd = pd;
+
+               if (cpumask_test_cpu(cpu, cpumask)
+                   && cpumask_test_cpu(cpu, cpu_active_mask)) {
+                       queue->cpu_index = cpu_index;
+                       cpu_index++;
+               } else
+                       queue->cpu_index = -1;
+
+               INIT_LIST_HEAD(&queue->reorder.list);
+               INIT_LIST_HEAD(&queue->parallel.list);
+               INIT_LIST_HEAD(&queue->serial.list);
+               spin_lock_init(&queue->reorder.lock);
+               spin_lock_init(&queue->parallel.lock);
+               spin_lock_init(&queue->serial.lock);
+
+               INIT_WORK(&queue->pwork, padata_parallel_worker);
+               INIT_WORK(&queue->swork, padata_serial_worker);
+               atomic_set(&queue->num_obj, 0);
+       }
+
+       cpumask_and(pd->cpumask, cpumask, cpu_active_mask);
+
+       num_cpus = cpumask_weight(pd->cpumask);
+       pd->max_seq_nr = (MAX_SEQ_NR / num_cpus) * num_cpus - 1;
+
+       atomic_set(&pd->seq_nr, -1);
+       atomic_set(&pd->reorder_objects, 0);
+       atomic_set(&pd->refcnt, 0);
+       pd->pinst = pinst;
+       spin_lock_init(&pd->lock);
+
+       return pd;
+
+err_free_queue:
+       free_percpu(pd->queue);
+err_free_pd:
+       kfree(pd);
+err:
+       return NULL;
+}
+
+static void padata_free_pd(struct parallel_data *pd)
+{
+       free_cpumask_var(pd->cpumask);
+       free_percpu(pd->queue);
+       kfree(pd);
+}
+
+static void padata_replace(struct padata_instance *pinst,
+                          struct parallel_data *pd_new)
+{
+       struct parallel_data *pd_old = pinst->pd;
+
+       pinst->flags |= PADATA_RESET;
+
+       rcu_assign_pointer(pinst->pd, pd_new);
+
+       synchronize_rcu();
+
+       while (atomic_read(&pd_old->refcnt) != 0)
+               yield();
+
+       flush_workqueue(pinst->wq);
+
+       padata_free_pd(pd_old);
+
+       pinst->flags &= ~PADATA_RESET;
+}
+
+/*
+ * padata_set_cpumask - set the cpumask that padata should use
+ *
+ * @pinst: padata instance
+ * @cpumask: the cpumask to use
+ */
+int padata_set_cpumask(struct padata_instance *pinst,
+                       cpumask_var_t cpumask)
+{
+       struct parallel_data *pd;
+       int err = 0;
+
+       might_sleep();
+
+       mutex_lock(&pinst->lock);
+
+       pd = padata_alloc_pd(pinst, cpumask);
+       if (!pd) {
+               err = -ENOMEM;
+               goto out;
+       }
+
+       cpumask_copy(pinst->cpumask, cpumask);
+
+       padata_replace(pinst, pd);
+
+out:
+       mutex_unlock(&pinst->lock);
+
+       return err;
+}
+EXPORT_SYMBOL(padata_set_cpumask);
+
+static int __padata_add_cpu(struct padata_instance *pinst, int cpu)
+{
+       struct parallel_data *pd;
+
+       if (cpumask_test_cpu(cpu, cpu_active_mask)) {
+               pd = padata_alloc_pd(pinst, pinst->cpumask);
+               if (!pd)
+                       return -ENOMEM;
+
+               padata_replace(pinst, pd);
+       }
+
+       return 0;
+}
+
+/*
+ * padata_add_cpu - add a cpu to the padata cpumask
+ *
+ * @pinst: padata instance
+ * @cpu: cpu to add
+ */
+int padata_add_cpu(struct padata_instance *pinst, int cpu)
+{
+       int err;
+
+       might_sleep();
+
+       mutex_lock(&pinst->lock);
+
+       cpumask_set_cpu(cpu, pinst->cpumask);
+       err = __padata_add_cpu(pinst, cpu);
+
+       mutex_unlock(&pinst->lock);
+
+       return err;
+}
+EXPORT_SYMBOL(padata_add_cpu);
+
+static int __padata_remove_cpu(struct padata_instance *pinst, int cpu)
+{
+       struct parallel_data *pd;
+
+       if (cpumask_test_cpu(cpu, cpu_online_mask)) {
+               pd = padata_alloc_pd(pinst, pinst->cpumask);
+               if (!pd)
+                       return -ENOMEM;
+
+               padata_replace(pinst, pd);
+       }
+
+       return 0;
+}
+
+/*
+ * padata_remove_cpu - remove a cpu from the padata cpumask
+ *
+ * @pinst: padata instance
+ * @cpu: cpu to remove
+ */
+int padata_remove_cpu(struct padata_instance *pinst, int cpu)
+{
+       int err;
+
+       might_sleep();
+
+       mutex_lock(&pinst->lock);
+
+       cpumask_clear_cpu(cpu, pinst->cpumask);
+       err = __padata_remove_cpu(pinst, cpu);
+
+       mutex_unlock(&pinst->lock);
+
+       return err;
+}
+EXPORT_SYMBOL(padata_remove_cpu);
+
+/*
+ * padata_start - start the parallel processing
+ *
+ * @pinst: padata instance to start
+ */
+void padata_start(struct padata_instance *pinst)
+{
+       might_sleep();
+
+       mutex_lock(&pinst->lock);
+       pinst->flags |= PADATA_INIT;
+       mutex_unlock(&pinst->lock);
+}
+EXPORT_SYMBOL(padata_start);
+
+/*
+ * padata_stop - stop the parallel processing
+ *
+ * @pinst: padata instance to stop
+ */
+void padata_stop(struct padata_instance *pinst)
+{
+       might_sleep();
+
+       mutex_lock(&pinst->lock);
+       pinst->flags &= ~PADATA_INIT;
+       mutex_unlock(&pinst->lock);
+}
+EXPORT_SYMBOL(padata_stop);
+
+static int __cpuinit padata_cpu_callback(struct notifier_block *nfb,
+                                        unsigned long action, void *hcpu)
+{
+       int err;
+       struct padata_instance *pinst;
+       int cpu = (unsigned long)hcpu;
+
+       pinst = container_of(nfb, struct padata_instance, cpu_notifier);
+
+       switch (action) {
+       case CPU_ONLINE:
+       case CPU_ONLINE_FROZEN:
+               if (!cpumask_test_cpu(cpu, pinst->cpumask))
+                       break;
+               mutex_lock(&pinst->lock);
+               err = __padata_add_cpu(pinst, cpu);
+               mutex_unlock(&pinst->lock);
+               if (err)
+                       return NOTIFY_BAD;
+               break;
+
+       case CPU_DOWN_PREPARE:
+       case CPU_DOWN_PREPARE_FROZEN:
+               if (!cpumask_test_cpu(cpu, pinst->cpumask))
+                       break;
+               mutex_lock(&pinst->lock);
+               err = __padata_remove_cpu(pinst, cpu);
+               mutex_unlock(&pinst->lock);
+               if (err)
+                       return NOTIFY_BAD;
+               break;
+
+       case CPU_UP_CANCELED:
+       case CPU_UP_CANCELED_FROZEN:
+               if (!cpumask_test_cpu(cpu, pinst->cpumask))
+                       break;
+               mutex_lock(&pinst->lock);
+               __padata_remove_cpu(pinst, cpu);
+               mutex_unlock(&pinst->lock);
+
+       case CPU_DOWN_FAILED:
+       case CPU_DOWN_FAILED_FROZEN:
+               if (!cpumask_test_cpu(cpu, pinst->cpumask))
+                       break;
+               mutex_lock(&pinst->lock);
+               __padata_add_cpu(pinst, cpu);
+               mutex_unlock(&pinst->lock);
+       }
+
+       return NOTIFY_OK;
+}
+
+/*
+ * padata_alloc - allocate and initialize a padata instance
+ *
+ * @cpumask: cpumask that padata uses for parallelization
+ * @wq: workqueue to use for the allocated padata instance
+ */
+struct padata_instance *padata_alloc(const struct cpumask *cpumask,
+                                    struct workqueue_struct *wq)
+{
+       int err;
+       struct padata_instance *pinst;
+       struct parallel_data *pd;
+
+       pinst = kzalloc(sizeof(struct padata_instance), GFP_KERNEL);
+       if (!pinst)
+               goto err;
+
+       pd = padata_alloc_pd(pinst, cpumask);
+       if (!pd)
+               goto err_free_inst;
+
+       rcu_assign_pointer(pinst->pd, pd);
+
+       pinst->wq = wq;
+
+       cpumask_copy(pinst->cpumask, cpumask);
+
+       pinst->flags = 0;
+
+       pinst->cpu_notifier.notifier_call = padata_cpu_callback;
+       pinst->cpu_notifier.priority = 0;
+       err = register_hotcpu_notifier(&pinst->cpu_notifier);
+       if (err)
+               goto err_free_pd;
+
+       mutex_init(&pinst->lock);
+
+       return pinst;
+
+err_free_pd:
+       padata_free_pd(pd);
+err_free_inst:
+       kfree(pinst);
+err:
+       return NULL;
+}
+EXPORT_SYMBOL(padata_alloc);
+
+/*
+ * padata_free - free a padata instance
+ *
+ * @ padata_inst: padata instance to free
+ */
+void padata_free(struct padata_instance *pinst)
+{
+       padata_stop(pinst);
+
+       synchronize_rcu();
+
+       while (atomic_read(&pinst->pd->refcnt) != 0)
+               yield();
+
+       unregister_hotcpu_notifier(&pinst->cpu_notifier);
+       padata_free_pd(pinst->pd);
+       kfree(pinst);
+}
+EXPORT_SYMBOL(padata_free);