#include <linux/namei.h>
#include <linux/pagemap.h>
#include <linux/proc_fs.h>
+#include <linux/rcupdate.h>
#include <linux/sched.h>
#include <linux/seq_file.h>
#include <linux/slab.h>
* When there is only one cpuset (the root cpuset) we can
* short circuit some hooks.
*/
-int number_of_cpusets;
+int number_of_cpusets __read_mostly;
/* See "Frequency meter" comments, below. */
* a tasks cpuset pointer we use task_lock(), which acts on a spinlock
* (task->alloc_lock) already in the task_struct routinely used for
* such matters.
+ *
+ * P.S. One more locking exception. RCU is used to guard the
+ * update of a tasks cpuset pointer by attach_task() and the
+ * access of task->cpuset->mems_generation via that pointer in
+ * the routine cpuset_update_task_memory_state().
*/
static DECLARE_MUTEX(manage_sem);
spin_lock(&dcache_lock);
node = dentry->d_subdirs.next;
while (node != &dentry->d_subdirs) {
- struct dentry *d = list_entry(node, struct dentry, d_child);
+ struct dentry *d = list_entry(node, struct dentry, d_u.d_child);
list_del_init(node);
if (d->d_inode) {
d = dget_locked(d);
}
node = dentry->d_subdirs.next;
}
- list_del_init(&dentry->d_child);
+ list_del_init(&dentry->d_u.d_child);
spin_unlock(&dcache_lock);
remove_dir(dentry);
}
* cpuset pointer. This routine also might acquire callback_sem and
* current->mm->mmap_sem during call.
*
- * The task_lock() is required to dereference current->cpuset safely.
- * Without it, we could pick up the pointer value of current->cpuset
- * in one instruction, and then attach_task could give us a different
- * cpuset, and then the cpuset we had could be removed and freed,
- * and then on our next instruction, we could dereference a no longer
- * valid cpuset pointer to get its mems_generation field.
+ * Reading current->cpuset->mems_generation doesn't need task_lock
+ * to guard the current->cpuset derefence, because it is guarded
+ * from concurrent freeing of current->cpuset by attach_task(),
+ * using RCU.
+ *
+ * The rcu_dereference() is technically probably not needed,
+ * as I don't actually mind if I see a new cpuset pointer but
+ * an old value of mems_generation. However this really only
+ * matters on alpha systems using cpusets heavily. If I dropped
+ * that rcu_dereference(), it would save them a memory barrier.
+ * For all other arch's, rcu_dereference is a no-op anyway, and for
+ * alpha systems not using cpusets, another planned optimization,
+ * avoiding the rcu critical section for tasks in the root cpuset
+ * which is statically allocated, so can't vanish, will make this
+ * irrelevant. Better to use RCU as intended, than to engage in
+ * some cute trick to save a memory barrier that is impossible to
+ * test, for alpha systems using cpusets heavily, which might not
+ * even exist.
*
* This routine is needed to update the per-task mems_allowed data,
* within the tasks context, when it is trying to allocate memory
* task has been modifying its cpuset.
*/
-void cpuset_update_task_memory_state()
+void cpuset_update_task_memory_state(void)
{
int my_cpusets_mem_gen;
struct task_struct *tsk = current;
- struct cpuset *cs = tsk->cpuset;
+ struct cpuset *cs;
- task_lock(tsk);
- my_cpusets_mem_gen = cs->mems_generation;
- task_unlock(tsk);
+ if (tsk->cpuset == &top_cpuset) {
+ /* Don't need rcu for top_cpuset. It's never freed. */
+ my_cpusets_mem_gen = top_cpuset.mems_generation;
+ } else {
+ rcu_read_lock();
+ cs = rcu_dereference(tsk->cpuset);
+ my_cpusets_mem_gen = cs->mems_generation;
+ rcu_read_unlock();
+ }
if (my_cpusets_mem_gen != tsk->cpuset_mems_generation) {
down(&callback_sem);
return -ESRCH;
}
atomic_inc(&cs->count);
- tsk->cpuset = cs;
+ rcu_assign_pointer(tsk->cpuset, cs);
task_unlock(tsk);
guarantee_online_cpus(cs, &cpus);
if (is_memory_migrate(cs))
do_migrate_pages(tsk->mm, &from, &to, MPOL_MF_MOVE_ALL);
put_task_struct(tsk);
+ synchronize_rcu();
if (atomic_dec_and_test(&oldcs->count))
check_for_release(oldcs, ppathbuf);
return 0;
struct dentry *dentry;
int error;
- down(&dir->d_inode->i_sem);
+ mutex_lock(&dir->d_inode->i_mutex);
dentry = cpuset_get_dentry(dir, cft->name);
if (!IS_ERR(dentry)) {
error = cpuset_create_file(dentry, 0644 | S_IFREG);
dput(dentry);
} else
error = PTR_ERR(dentry);
- up(&dir->d_inode->i_sem);
+ mutex_unlock(&dir->d_inode->i_mutex);
return error;
}
* when reading out p->cpuset, as we don't really care if it changes
* on the next cycle, and we are not going to try to dereference it.
*/
-static inline int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs)
+static int pid_array_load(pid_t *pidarray, int npids, struct cpuset *cs)
{
int n = 0;
struct task_struct *g, *p;
/*
* Release manage_sem before cpuset_populate_dir() because it
- * will down() this new directory's i_sem and if we race with
+ * will down() this new directory's i_mutex and if we race with
* another mkdir, we might deadlock.
*/
up(&manage_sem);
{
struct cpuset *c_parent = dentry->d_parent->d_fsdata;
- /* the vfs holds inode->i_sem already */
+ /* the vfs holds inode->i_mutex already */
return cpuset_create(c_parent, dentry->d_name.name, mode | S_IFDIR);
}
struct cpuset *parent;
char *pathbuf = NULL;
- /* the vfs holds both inode->i_sem already */
+ /* the vfs holds both inode->i_mutex already */
down(&manage_sem);
cpuset_update_task_memory_state();
* We don't need to task_lock() this reference to tsk->cpuset,
* because tsk is already marked PF_EXITING, so attach_task() won't
* mess with it, or task is a failed fork, never visible to attach_task.
+ *
+ * Hack:
+ *
+ * Set the exiting tasks cpuset to the root cpuset (top_cpuset).
+ *
+ * Don't leave a task unable to allocate memory, as that is an
+ * accident waiting to happen should someone add a callout in
+ * do_exit() after the cpuset_exit() call that might allocate.
+ * If a task tries to allocate memory with an invalid cpuset,
+ * it will oops in cpuset_update_task_memory_state().
+ *
+ * We call cpuset_exit() while the task is still competent to
+ * handle notify_on_release(), then leave the task attached to
+ * the root cpuset (top_cpuset) for the remainder of its exit.
+ *
+ * To do this properly, we would increment the reference count on
+ * top_cpuset, and near the very end of the kernel/exit.c do_exit()
+ * code we would add a second cpuset function call, to drop that
+ * reference. This would just create an unnecessary hot spot on
+ * the top_cpuset reference count, to no avail.
+ *
+ * Normally, holding a reference to a cpuset without bumping its
+ * count is unsafe. The cpuset could go away, or someone could
+ * attach us to a different cpuset, decrementing the count on
+ * the first cpuset that we never incremented. But in this case,
+ * top_cpuset isn't going away, and either task has PF_EXITING set,
+ * which wards off any attach_task() attempts, or task is a failed
+ * fork, never visible to attach_task.
+ *
+ * Another way to do this would be to set the cpuset pointer
+ * to NULL here, and check in cpuset_update_task_memory_state()
+ * for a NULL pointer. This hack avoids that NULL check, for no
+ * cost (other than this way too long comment ;).
**/
void cpuset_exit(struct task_struct *tsk)
struct cpuset *cs;
cs = tsk->cpuset;
- tsk->cpuset = NULL;
+ tsk->cpuset = &top_cpuset; /* Hack - see comment above */
if (notify_on_release(cs)) {
char *pathbuf = NULL;
return allowed;
}
+/**
+ * cpuset_lock - lock out any changes to cpuset structures
+ *
+ * The out of memory (oom) code needs to lock down cpusets
+ * from being changed while it scans the tasklist looking for a
+ * task in an overlapping cpuset. Expose callback_sem via this
+ * cpuset_lock() routine, so the oom code can lock it, before
+ * locking the task list. The tasklist_lock is a spinlock, so
+ * must be taken inside callback_sem.
+ */
+
+void cpuset_lock(void)
+{
+ down(&callback_sem);
+}
+
+/**
+ * cpuset_unlock - release lock on cpuset changes
+ *
+ * Undo the lock taken in a previous cpuset_lock() call.
+ */
+
+void cpuset_unlock(void)
+{
+ up(&callback_sem);
+}
+
/**
* cpuset_excl_nodes_overlap - Do we overlap @p's mem_exclusive ancestors?
* @p: pointer to task_struct of some other task.
* determine if task @p's memory usage might impact the memory
* available to the current task.
*
- * Acquires callback_sem - not suitable for calling from a fast path.
+ * Call while holding callback_sem.
**/
int cpuset_excl_nodes_overlap(const struct task_struct *p)
const struct cpuset *cs1, *cs2; /* my and p's cpuset ancestors */
int overlap = 0; /* do cpusets overlap? */
- down(&callback_sem);
-
task_lock(current);
if (current->flags & PF_EXITING) {
task_unlock(current);
overlap = nodes_intersects(cs1->mems_allowed, cs2->mems_allowed);
done:
- up(&callback_sem);
-
return overlap;
}