fix acct_collect() siglock bug
PATCH 1/1 fix acct_collect() siglock bug
the patch below fixes a siglock usage bug introduced in -mm2. The lock validator found it during bootup.
---------------
Subject: fix acct_collect() siglock bug
From: Ingo Molnar <[email protected]>
the lock validator noticed the following bug:
=========================================================
[ INFO: possible irq lock inversion dependency detected ]
---------------------------------------------------------
swapper/0 just changed the state of lock:
(tasklist_lock){..-?}, at: [<c01247c8>] send_group_sig_info+0xe/0x2f
but this lock took another, soft-irq-unsafe lock in the past:
(&sighand->siglock){--..}
and interrupts could create inverse lock ordering between them.
...
-> (init_sighand.siglock){....} ops: 18 {
...
hardirq-on-W at:
[<c0130577>] lock_acquire+0x56/0x73
[<c03ec59c>] _spin_lock+0x24/0x32
[<c01374bd>] acct_collect+0x81/0x115
[<c011b4fa>] do_exit+0x19a/0x7b3
[<c01273ce>] ____call_usermodehelper+0xdf/0xe1
[<c01007f1>] kernel_thread_helper+0x5/0xbit turns out that acct_collect() takes current->sighand->siglock with interrupts enabled - a big no-no. Fix it by disabling interrupts.
--- kernel/acct.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
kernel/acct.c +2 -2
@@ -549,7 +549,7 @@ void acct_collect(long exitcode, int group_read (¤t->mm->mmap_sem); } -spin_lock (¤t->sighand->siglock); +spin_lock_irq (¤t->sighand->siglock);if (group_dead) pacct->ac_mem = vsize /1024 ;if (thread_group_leader (current)) {
@@ -567,7 +567,7 @@ void acct_collect(long exitcode, int gro pacct->ac_stime =cputime_add (pacct->ac_stime, current->stime); pacct->ac_minflt += current->min_flt; pacct->ac_majflt += current->maj_flt; -spin_unlock (¤t->sighand->siglock); +spin_unlock_irq (¤t->sighand->siglock); }/**