Merge branch 'urgent-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu

Pull RCU fix from Paul McKenney:
 "A single commit that fixes a bug that was introduced a couple of merge
  windows ago, but which rather more recently converged to an
  agreed-upon fix. The bug is that interrupts can be incorrectly enabled
  while holding an irq-disabled spinlock. This can of course result in
  self-deadlocks.

  The bug is a bit difficult to trigger. It requires that a preempted
  task be blocking a preemptible-RCU grace period long enough to trigger
  an RCU CPU stall warning. In addition, an interrupt must occur at just
  the right time, and that interrupt's handler must acquire that same
  irq-disabled spinlock. Still, a deadlock is a deadlock.

  Furthermore, we do now have a fix, and that fix survives kernel test
  robot, -next, and rcutorture testing. It has also been verified by
  Sebastian as fixing the bug. Therefore..."

* 'urgent-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu:
  rcu: Don't invoke try_invoke_on_locked_down_task() with irqs disabled
This commit is contained in:
Linus Torvalds 2020-11-17 10:31:56 -08:00
commit 9dacf44c38

View File

@ -249,13 +249,16 @@ static bool check_slow_task(struct task_struct *t, void *arg)
/*
* Scan the current list of tasks blocked within RCU read-side critical
* sections, printing out the tid of each.
* sections, printing out the tid of each of the first few of them.
*/
static int rcu_print_task_stall(struct rcu_node *rnp)
static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
__releases(rnp->lock)
{
int i = 0;
int ndetected = 0;
struct rcu_stall_chk_rdr rscr;
struct task_struct *t;
struct task_struct *ts[8];
if (!rcu_preempt_blocked_readers_cgp(rnp))
return 0;
@ -264,6 +267,14 @@ static int rcu_print_task_stall(struct rcu_node *rnp)
t = list_entry(rnp->gp_tasks->prev,
struct task_struct, rcu_node_entry);
list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) {
get_task_struct(t);
ts[i++] = t;
if (i >= ARRAY_SIZE(ts))
break;
}
raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
for (i--; i; i--) {
t = ts[i];
if (!try_invoke_on_locked_down_task(t, check_slow_task, &rscr))
pr_cont(" P%d", t->pid);
else
@ -273,6 +284,7 @@ static int rcu_print_task_stall(struct rcu_node *rnp)
".q"[rscr.rs.b.need_qs],
".e"[rscr.rs.b.exp_hint],
".l"[rscr.on_blkd_list]);
put_task_struct(t);
ndetected++;
}
pr_cont("\n");
@ -293,8 +305,9 @@ static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
* Because preemptible RCU does not exist, we never have to check for
* tasks blocked within RCU read-side critical sections.
*/
static int rcu_print_task_stall(struct rcu_node *rnp)
static int rcu_print_task_stall(struct rcu_node *rnp, unsigned long flags)
{
raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
return 0;
}
#endif /* #else #ifdef CONFIG_PREEMPT_RCU */
@ -472,7 +485,6 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
pr_err("INFO: %s detected stalls on CPUs/tasks:\n", rcu_state.name);
rcu_for_each_leaf_node(rnp) {
raw_spin_lock_irqsave_rcu_node(rnp, flags);
ndetected += rcu_print_task_stall(rnp);
if (rnp->qsmask != 0) {
for_each_leaf_node_possible_cpu(rnp, cpu)
if (rnp->qsmask & leaf_node_cpu_bit(rnp, cpu)) {
@ -480,7 +492,7 @@ static void print_other_cpu_stall(unsigned long gp_seq, unsigned long gps)
ndetected++;
}
}
raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
ndetected += rcu_print_task_stall(rnp, flags); // Releases rnp->lock.
}
for_each_possible_cpu(cpu)