Re: fs/exec.c:1307:26: sparse: sparse: incorrect type in argument 1 (different address spaces)

From: Thomas Gleixner
Date: Fri Jan 12 2024 - 10:32:36 EST


On Thu, Jan 11 2024 at 10:44, kernel test robot wrote:
> tree: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> head: acc657692aed438e9931438f8c923b2b107aebf9
> commit: e362359ace6f87c201531872486ff295df306d13 posix-cpu-timers: Cleanup CPU timers before freeing them during exec
> date: 1 year, 5 months ago

I'm amused that it took 17 month ....

>>> fs/exec.c:1307:26: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected struct spinlock [usertype] *lock @@ got struct spinlock [noderef] __rcu * @@
> 1305
> 1306 #ifdef CONFIG_POSIX_TIMERS
>> 1307 spin_lock_irq(&me->sighand->siglock);
> 1308 posix_cpu_timers_exit(me);
> 1309 spin_unlock_irq(&me->sighand->siglock);
> 1310 exit_itimers(me);
> 1311 flush_itimer_signals();
> 1312 #endif

So this warning is clearly wrong because 'current->sighand->siglock' is
safe to dereference w/o RCU protection for 'current'.

The real issue is this commit:

913292c97d75 ("sched.h: Annotate sighand_struct with __rcu")

which blindly 'fixed' a sparse warning in signal.c w/o even trying to
look at the consequences. There are 170+ instances of spin_[un]lock()
variants which should emit exactly the same warning...

I think the right fix is to annotate this legit case of derefencing
current->sighand->siglock so sparse knows that this is safe.

Thanks,

tglx