padata: Fix race on sequence number wrap
authorSteffen Klassert <steffen.klassert@secunet.com>
Fri, 9 Mar 2012 06:20:49 +0000 (07:20 +0100)
committerHerbert Xu <herbert@gondor.apana.org.au>
Wed, 14 Mar 2012 09:25:56 +0000 (17:25 +0800)
When padata_do_parallel() is called from multiple cpus for the same
padata instance, we can get object reordering on sequence number wrap
because testing for sequence number wrap and reseting the sequence
number must happen atomically but is implemented with two atomic
operations. This patch fixes this by converting the sequence number
from atomic_t to an unsigned int and protect the access with a
spin_lock. As a side effect, we get rid of the sequence number wrap
handling because the seqence number wraps back to null now without
the need to do anything.

Signed-off-by: Steffen Klassert <steffen.klassert@secunet.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
include/linux/padata.h
kernel/padata.c

index 4633b2f..86292be 100644 (file)
@@ -46,7 +46,6 @@ 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);
@@ -116,7 +115,6 @@ struct padata_cpumask {
  * @pinst: padata instance.
  * @pqueue: percpu padata queues used for parallelization.
  * @squeue: percpu padata queues used for serialuzation.
- * @seq_nr: The sequence number that will be attached to the next object.
  * @reorder_objects: Number of objects waiting in the reorder queues.
  * @refcnt: Number of objects holding a reference on this parallel_data.
  * @max_seq_nr:  Maximal used sequence number.
@@ -129,12 +127,12 @@ struct parallel_data {
        struct padata_instance          *pinst;
        struct padata_parallel_queue    __percpu *pqueue;
        struct padata_serial_queue      __percpu *squeue;
-       atomic_t                        seq_nr;
        atomic_t                        reorder_objects;
        atomic_t                        refcnt;
-       unsigned int                    max_seq_nr;
        struct padata_cpumask           cpumask;
        spinlock_t                      lock ____cacheline_aligned;
+       spinlock_t                      seq_lock;
+       unsigned int                    seq_nr;
        unsigned int                    processed;
        struct timer_list               timer;
 };
index aa99295..6f10eb2 100644 (file)
@@ -29,7 +29,6 @@
 #include <linux/sysfs.h>
 #include <linux/rcupdate.h>
 
-#define MAX_SEQ_NR (INT_MAX - NR_CPUS)
 #define MAX_OBJ_NUM 1000
 
 static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
@@ -43,18 +42,19 @@ static int padata_index_to_cpu(struct parallel_data *pd, int cpu_index)
        return target_cpu;
 }
 
-static int padata_cpu_hash(struct padata_priv *padata)
+static int padata_cpu_hash(struct parallel_data *pd)
 {
        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.pcpu);
+
+       spin_lock(&pd->seq_lock);
+       cpu_index =  pd->seq_nr % cpumask_weight(pd->cpumask.pcpu);
+       pd->seq_nr++;
+       spin_unlock(&pd->seq_lock);
 
        return padata_index_to_cpu(pd, cpu_index);
 }
@@ -132,12 +132,7 @@ int padata_do_parallel(struct padata_instance *pinst,
        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);
+       target_cpu = padata_cpu_hash(pd);
        queue = per_cpu_ptr(pd->pqueue, target_cpu);
 
        spin_lock(&queue->parallel.lock);
@@ -173,7 +168,7 @@ EXPORT_SYMBOL(padata_do_parallel);
 static struct padata_priv *padata_get_next(struct parallel_data *pd)
 {
        int cpu, num_cpus;
-       int next_nr, next_index;
+       unsigned int next_nr, next_index;
        struct padata_parallel_queue *queue, *next_queue;
        struct padata_priv *padata;
        struct padata_list *reorder;
@@ -189,14 +184,6 @@ static struct padata_priv *padata_get_next(struct parallel_data *pd)
        cpu = padata_index_to_cpu(pd, next_index);
        next_queue = per_cpu_ptr(pd->pqueue, cpu);
 
-       if (unlikely(next_nr > pd->max_seq_nr)) {
-               next_nr = next_nr - pd->max_seq_nr - 1;
-               next_index = next_nr % num_cpus;
-               cpu = padata_index_to_cpu(pd, next_index);
-               next_queue = per_cpu_ptr(pd->pqueue, cpu);
-               pd->processed = 0;
-       }
-
        padata = NULL;
 
        reorder = &next_queue->reorder;
@@ -205,8 +192,6 @@ static struct padata_priv *padata_get_next(struct parallel_data *pd)
                padata = list_entry(reorder->list.next,
                                    struct padata_priv, list);
 
-               BUG_ON(next_nr != padata->seq_nr);
-
                spin_lock(&reorder->lock);
                list_del_init(&padata->list);
                atomic_dec(&pd->reorder_objects);
@@ -402,7 +387,7 @@ static void padata_init_squeues(struct parallel_data *pd)
 /* Initialize all percpu queues used by parallel workers */
 static void padata_init_pqueues(struct parallel_data *pd)
 {
-       int cpu_index, num_cpus, cpu;
+       int cpu_index, cpu;
        struct padata_parallel_queue *pqueue;
 
        cpu_index = 0;
@@ -417,9 +402,6 @@ static void padata_init_pqueues(struct parallel_data *pd)
                INIT_WORK(&pqueue->work, padata_parallel_worker);
                atomic_set(&pqueue->num_obj, 0);
        }
-
-       num_cpus = cpumask_weight(pd->cpumask.pcpu);
-       pd->max_seq_nr = num_cpus ? (MAX_SEQ_NR / num_cpus) * num_cpus - 1 : 0;
 }
 
 /* Allocate and initialize the internal cpumask dependend resources. */
@@ -446,7 +428,7 @@ static struct parallel_data *padata_alloc_pd(struct padata_instance *pinst,
        padata_init_pqueues(pd);
        padata_init_squeues(pd);
        setup_timer(&pd->timer, padata_reorder_timer, (unsigned long)pd);
-       atomic_set(&pd->seq_nr, -1);
+       pd->seq_nr = 0;
        atomic_set(&pd->reorder_objects, 0);
        atomic_set(&pd->refcnt, 0);
        pd->pinst = pinst;