slow-work: use round_jiffies() for thread pool's cull and OOM timers
authorChris Peterson <cpeterso@cpeterso.com>
Tue, 16 Jun 2009 22:33:43 +0000 (15:33 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Wed, 17 Jun 2009 02:47:49 +0000 (19:47 -0700)
Round the slow work queue's cull and OOM timeouts to whole second boundary
with round_jiffies().  The slow work queue uses a pair of timers to cull
idle threads and, after OOM, to delay new thread creation.

This patch also extracts the mod_timer() logic for the cull timer into a
separate helper function.

By rounding non-time-critical timers such as these to whole seconds, they
will be batched up to fire at the same time rather than being spread out.
This allows the CPU wake up less, which saves power.

Signed-off-by: Chris Peterson <cpeterso@cpeterso.com>
Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/slow-work.c

index 521ed2004d63fc7cf2bfbd8eff33d7d1bf58f1d3..09d7519557d35181fab35591a17516efe5aba723 100644 (file)
@@ -318,6 +318,15 @@ cant_get_ref:
 }
 EXPORT_SYMBOL(slow_work_enqueue);
 
+/*
+ * Schedule a cull of the thread pool at some time in the near future
+ */
+static void slow_work_schedule_cull(void)
+{
+       mod_timer(&slow_work_cull_timer,
+                 round_jiffies(jiffies + SLOW_WORK_CULL_TIMEOUT));
+}
+
 /*
  * Worker thread culling algorithm
  */
@@ -335,8 +344,7 @@ static bool slow_work_cull_thread(void)
                    list_empty(&vslow_work_queue) &&
                    atomic_read(&slow_work_thread_count) >
                    slow_work_min_threads) {
-                       mod_timer(&slow_work_cull_timer,
-                                 jiffies + SLOW_WORK_CULL_TIMEOUT);
+                       slow_work_schedule_cull();
                        do_cull = true;
                }
        }
@@ -393,8 +401,7 @@ static int slow_work_thread(void *_data)
                            list_empty(&vslow_work_queue) &&
                            atomic_read(&slow_work_thread_count) >
                            slow_work_min_threads)
-                               mod_timer(&slow_work_cull_timer,
-                                         jiffies + SLOW_WORK_CULL_TIMEOUT);
+                               slow_work_schedule_cull();
                        continue;
                }
 
@@ -458,7 +465,7 @@ static void slow_work_new_thread_execute(struct slow_work *work)
                if (atomic_dec_and_test(&slow_work_thread_count))
                        BUG(); /* we're running on a slow work thread... */
                mod_timer(&slow_work_oom_timer,
-                         jiffies + SLOW_WORK_OOM_TIMEOUT);
+                         round_jiffies(jiffies + SLOW_WORK_OOM_TIMEOUT));
        } else {
                /* ratelimit the starting of new threads */
                mod_timer(&slow_work_oom_timer, jiffies + 1);
@@ -502,8 +509,7 @@ static int slow_work_min_threads_sysctl(struct ctl_table *table, int write,
                        if (n < 0 && !slow_work_may_not_start_new_thread)
                                slow_work_enqueue(&slow_work_new_thread);
                        else if (n > 0)
-                               mod_timer(&slow_work_cull_timer,
-                                         jiffies + SLOW_WORK_CULL_TIMEOUT);
+                               slow_work_schedule_cull();
                }
                mutex_unlock(&slow_work_user_lock);
        }
@@ -529,8 +535,7 @@ static int slow_work_max_threads_sysctl(struct ctl_table *table, int write,
                                atomic_read(&slow_work_thread_count);
 
                        if (n < 0)
-                               mod_timer(&slow_work_cull_timer,
-                                         jiffies + SLOW_WORK_CULL_TIMEOUT);
+                               slow_work_schedule_cull();
                }
                mutex_unlock(&slow_work_user_lock);
        }