]> err.no Git - linux-2.6/commitdiff
cpusets: update_cpumask revision
authorCliff Wickman <cpw@sgi.com>
Thu, 7 Feb 2008 08:14:44 +0000 (00:14 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 7 Feb 2008 16:42:22 +0000 (08:42 -0800)
Use the new function cgroup_scan_tasks() to step through all tasks in a
cpuset.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: Cliff Wickman <cpw@sgi.com>
Cc: Paul Menage <menage@google.com>
Cc: Paul Jackson <pj@sgi.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
kernel/cpuset.c

index d94a8f7c4c294454366ea4bdc116fbdee7f3f840..20cb3916c66cb7e2de5555228b78ccdbf4ede2d9 100644 (file)
@@ -38,7 +38,6 @@
 #include <linux/mount.h>
 #include <linux/namei.h>
 #include <linux/pagemap.h>
-#include <linux/prio_heap.h>
 #include <linux/proc_fs.h>
 #include <linux/rcupdate.h>
 #include <linux/sched.h>
@@ -740,22 +739,50 @@ static inline int started_after(void *p1, void *p2)
        return started_after_time(t1, &t2->start_time, t2);
 }
 
-/*
+/**
+ * cpuset_test_cpumask - test a task's cpus_allowed versus its cpuset's
+ * @tsk: task to test
+ * @scan: struct cgroup_scanner contained in its struct cpuset_hotplug_scanner
+ *
  * Call with manage_mutex held.  May take callback_mutex during call.
+ * Called for each task in a cgroup by cgroup_scan_tasks().
+ * Return nonzero if this tasks's cpus_allowed mask should be changed (in other
+ * words, if its mask is not equal to its cpuset's mask).
  */
+int cpuset_test_cpumask(struct task_struct *tsk, struct cgroup_scanner *scan)
+{
+       return !cpus_equal(tsk->cpus_allowed,
+                       (cgroup_cs(scan->cg))->cpus_allowed);
+}
 
+/**
+ * cpuset_change_cpumask - make a task's cpus_allowed the same as its cpuset's
+ * @tsk: task to test
+ * @scan: struct cgroup_scanner containing the cgroup of the task
+ *
+ * Called by cgroup_scan_tasks() for each task in a cgroup whose
+ * cpus_allowed mask needs to be changed.
+ *
+ * We don't need to re-check for the cgroup/cpuset membership, since we're
+ * holding cgroup_lock() at this point.
+ */
+void cpuset_change_cpumask(struct task_struct *tsk, struct cgroup_scanner *scan)
+{
+       set_cpus_allowed(tsk, (cgroup_cs(scan->cg))->cpus_allowed);
+}
+
+/**
+ * update_cpumask - update the cpus_allowed mask of a cpuset and all tasks in it
+ * @cs: the cpuset to consider
+ * @buf: buffer of cpu numbers written to this cpuset
+ */
 static int update_cpumask(struct cpuset *cs, char *buf)
 {
        struct cpuset trialcs;
-       int retval, i;
-       int is_load_balanced;
-       struct cgroup_iter it;
-       struct cgroup *cgrp = cs->css.cgroup;
-       struct task_struct *p, *dropped;
-       /* Never dereference latest_task, since it's not refcounted */
-       struct task_struct *latest_task = NULL;
+       struct cgroup_scanner scan;
        struct ptr_heap heap;
-       struct timespec latest_time = { 0, 0 };
+       int retval;
+       int is_load_balanced;
 
        /* top_cpuset.cpus_allowed tracks cpu_online_map; it's read-only */
        if (cs == &top_cpuset)
@@ -764,7 +791,7 @@ static int update_cpumask(struct cpuset *cs, char *buf)
        trialcs = *cs;
 
        /*
-        * An empty cpus_allowed is ok iff there are no tasks in the cpuset.
+        * An empty cpus_allowed is ok if there are no tasks in the cpuset.
         * Since cpulist_parse() fails on an empty mask, we special case
         * that parsing.  The validate_change() call ensures that cpusets
         * with tasks have cpus.
@@ -785,6 +812,7 @@ static int update_cpumask(struct cpuset *cs, char *buf)
        /* Nothing to do if the cpus didn't change */
        if (cpus_equal(cs->cpus_allowed, trialcs.cpus_allowed))
                return 0;
+
        retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, &started_after);
        if (retval)
                return retval;
@@ -795,62 +823,19 @@ static int update_cpumask(struct cpuset *cs, char *buf)
        cs->cpus_allowed = trialcs.cpus_allowed;
        mutex_unlock(&callback_mutex);
 
- again:
        /*
         * Scan tasks in the cpuset, and update the cpumasks of any
-        * that need an update. Since we can't call set_cpus_allowed()
-        * while holding tasklist_lock, gather tasks to be processed
-        * in a heap structure. If the statically-sized heap fills up,
-        * overflow tasks that started later, and in future iterations
-        * only consider tasks that started after the latest task in
-        * the previous pass. This guarantees forward progress and
-        * that we don't miss any tasks
+        * that need an update.
         */
-       heap.size = 0;
-       cgroup_iter_start(cgrp, &it);
-       while ((p = cgroup_iter_next(cgrp, &it))) {
-               /* Only affect tasks that don't have the right cpus_allowed */
-               if (cpus_equal(p->cpus_allowed, cs->cpus_allowed))
-                       continue;
-               /*
-                * Only process tasks that started after the last task
-                * we processed
-                */
-               if (!started_after_time(p, &latest_time, latest_task))
-                       continue;
-               dropped = heap_insert(&heap, p);
-               if (dropped == NULL) {
-                       get_task_struct(p);
-               } else if (dropped != p) {
-                       get_task_struct(p);
-                       put_task_struct(dropped);
-               }
-       }
-       cgroup_iter_end(cgrp, &it);
-       if (heap.size) {
-               for (i = 0; i < heap.size; i++) {
-                       struct task_struct *p = heap.ptrs[i];
-                       if (i == 0) {
-                               latest_time = p->start_time;
-                               latest_task = p;
-                       }
-                       set_cpus_allowed(p, cs->cpus_allowed);
-                       put_task_struct(p);
-               }
-               /*
-                * If we had to process any tasks at all, scan again
-                * in case some of them were in the middle of forking
-                * children that didn't notice the new cpumask
-                * restriction.  Not the most efficient way to do it,
-                * but it avoids having to take callback_mutex in the
-                * fork path
-                */
-               goto again;
-       }
+       scan.cg = cs->css.cgroup;
+       scan.test_task = cpuset_test_cpumask;
+       scan.process_task = cpuset_change_cpumask;
+       scan.heap = &heap;
+       cgroup_scan_tasks(&scan);
        heap_free(&heap);
+
        if (is_load_balanced)
                rebuild_sched_domains();
-
        return 0;
 }