]> git.proxmox.com Git - mirror_ubuntu-kernels.git/commitdiff
workqueue: add workqueue_struct->pwqs list
authorTejun Heo <tj@kernel.org>
Tue, 12 Mar 2013 18:29:57 +0000 (11:29 -0700)
committerTejun Heo <tj@kernel.org>
Tue, 12 Mar 2013 18:29:57 +0000 (11:29 -0700)
Add workqueue_struct->pwqs list and chain all pool_workqueues
belonging to a workqueue there.  This will be used to implement
generic pool_workqueue iteration and handle multiple pool_workqueues
for the scheduled unbound pools with custom attributes.

This patch doesn't introduce any visible behavior changes.

Signed-off-by: Tejun Heo <tj@kernel.org>
Reviewed-by: Lai Jiangshan <laijs@cn.fujitsu.com>
kernel/workqueue.c

index f9e2ad9a320551a85dcaf024a0a35b1347312732..8634fc9d52d265a86385c4a0829fbcdaa0502065 100644 (file)
@@ -169,6 +169,7 @@ struct pool_workqueue {
        int                     nr_active;      /* L: nr of active works */
        int                     max_active;     /* L: max active works */
        struct list_head        delayed_works;  /* L: delayed works */
+       struct list_head        pwqs_node;      /* I: node on wq->pwqs */
 } __aligned(1 << WORK_STRUCT_FLAG_BITS);
 
 /*
@@ -212,6 +213,7 @@ struct workqueue_struct {
                struct pool_workqueue                   *single;
                unsigned long                           v;
        } pool_wq;                              /* I: pwq's */
+       struct list_head        pwqs;           /* I: all pwqs of this wq */
        struct list_head        list;           /* W: list of all workqueues */
 
        struct mutex            flush_mutex;    /* protects wq flushing */
@@ -3096,14 +3098,32 @@ int keventd_up(void)
        return system_wq != NULL;
 }
 
-static int alloc_pwqs(struct workqueue_struct *wq)
+static int alloc_and_link_pwqs(struct workqueue_struct *wq)
 {
-       if (!(wq->flags & WQ_UNBOUND))
+       int cpu;
+
+       if (!(wq->flags & WQ_UNBOUND)) {
                wq->pool_wq.pcpu = alloc_percpu(struct pool_workqueue);
-       else
-               wq->pool_wq.single = kmem_cache_zalloc(pwq_cache, GFP_KERNEL);
+               if (!wq->pool_wq.pcpu)
+                       return -ENOMEM;
+
+               for_each_possible_cpu(cpu) {
+                       struct pool_workqueue *pwq = get_pwq(cpu, wq);
 
-       return wq->pool_wq.v ? 0 : -ENOMEM;
+                       list_add_tail(&pwq->pwqs_node, &wq->pwqs);
+               }
+       } else {
+               struct pool_workqueue *pwq;
+
+               pwq = kmem_cache_zalloc(pwq_cache, GFP_KERNEL);
+               if (!pwq)
+                       return -ENOMEM;
+
+               wq->pool_wq.single = pwq;
+               list_add_tail(&pwq->pwqs_node, &wq->pwqs);
+       }
+
+       return 0;
 }
 
 static void free_pwqs(struct workqueue_struct *wq)
@@ -3165,13 +3185,14 @@ struct workqueue_struct *__alloc_workqueue_key(const char *fmt,
        wq->saved_max_active = max_active;
        mutex_init(&wq->flush_mutex);
        atomic_set(&wq->nr_pwqs_to_flush, 0);
+       INIT_LIST_HEAD(&wq->pwqs);
        INIT_LIST_HEAD(&wq->flusher_queue);
        INIT_LIST_HEAD(&wq->flusher_overflow);
 
        lockdep_init_map(&wq->lockdep_map, lock_name, key, 0);
        INIT_LIST_HEAD(&wq->list);
 
-       if (alloc_pwqs(wq) < 0)
+       if (alloc_and_link_pwqs(wq) < 0)
                goto err;
 
        for_each_pwq_cpu(cpu, wq) {