[PATCH] signal: annotate siglock acquisition

[PATCH] signal: annotate siglock acquisition

[Namhyung Kim]

lock_task_sighand() conditionally acquires sighand->siglock in case of
returning non-NULL but unlock_task_sighand() releases it unconditionally.
This leads sparse to complain about the lock context imbalance. Annotate it
to make sparse happier.

Signed-off-by: Namhyung Kim <namhyung@gmail.com>
---
 kernel/signal.c |    9 ++++++---
 1 files changed, 6 insertions(+), 3 deletions(-)

diff --git a/kernel/signal.c b/kernel/signal.c
index bded651..85c1efd 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -1037,7 +1037,7 @@ int do_send_sig_info(int sig, struct siginfo *info, struct task_struct *p,
 	unsigned long flags;
 	int ret = -ESRCH;
 
-	if (lock_task_sighand(p, &flags)) {
+	if (__cond_lock(&p->sighand->siglock, lock_task_sighand(p, &flags))) {
 		ret = send_signal(sig, info, p, group);
 		unlock_task_sighand(p, &flags);
 	}
@@ -1226,7 +1226,8 @@ int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
 		goto out_unlock;
 
 	if (sig) {
-		if (lock_task_sighand(p, &flags)) {
+		if (__cond_lock(&p->sighand->siglock,
+				lock_task_sighand(p, &flags))) {
 			ret = __send_signal(sig, info, p, 1, 0);
 			unlock_task_sighand(p, &flags);
 		} else
@@ -1403,7 +1404,9 @@ int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
 	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
 
 	ret = -1;
-	if (!likely(lock_task_sighand(t, &flags)))
+	if (likely(lock_task_sighand(t, &flags)))
+		__acquire(&t->sighand->siglock);
+	else
 		goto ret;
 
 	ret = 1; /* the signal is ignored */
-- 
1.7.2.2

Re: [PATCH] signal: annotate siglock acquisition

[Eric Dumazet]

Le samedi 25 septembre 2010 à 18:21 +0900, Namhyung Kim a écrit :
> lock_task_sighand() conditionally acquires sighand->siglock in case of
> returning non-NULL but unlock_task_sighand() releases it unconditionally.
> This leads sparse to complain about the lock context imbalance. Annotate it
> to make sparse happier.

> @@ -1403,7 +1404,9 @@ int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
>  	BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
>  
>  	ret = -1;
> -	if (!likely(lock_task_sighand(t, &flags)))
> +	if (likely(lock_task_sighand(t, &flags)))
> +		__acquire(&t->sighand->siglock);
> +	else
>  		goto ret;
>  
>  	ret = 1; /* the signal is ignored */

making sparse happy is good, but changing a likely clause is not,
unless you meant it ;)

maybe -->

	if (!likely(lock_task_sighand(t, &flags)))
		goto ret;
	else
		__acquire(&t->sighand->siglock);

Re: [PATCH] signal: annotate siglock acquisition

[Namhyung Kim]

On Sat, Sep 25, 2010 at 18:42, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> Le samedi 25 septembre 2010 à 18:21 +0900, Namhyung Kim a écrit :
>> @@ -1403,7 +1404,9 @@ int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
>>       BUG_ON(!(q->flags & SIGQUEUE_PREALLOC));
>>
>>       ret = -1;
>> -     if (!likely(lock_task_sighand(t, &flags)))
>> +     if (likely(lock_task_sighand(t, &flags)))
>> +             __acquire(&t->sighand->siglock);
>> +     else
>>               goto ret;
>>
>>       ret = 1; /* the signal is ignored */
>
> making sparse happy is good, but changing a likely clause is not,
> unless you meant it ;)
>
> maybe -->
>
>        if (!likely(lock_task_sighand(t, &flags)))
>                goto ret;
>        else
>                __acquire(&t->sighand->siglock);
>
>
>
>

What's the difference? I didn't change the semantics of likely clause
but simply exchange the order of then-else part because IMHO a
negative expression make it hard to understand, generally.
I've checked compiled binaries for both cases on x86 and they
produced exactly same code.

Thanks.


-- 
Regards,
Namhyung Kim

Re: [PATCH] signal: annotate siglock acquisition

[Eric Dumazet]

Le samedi 25 septembre 2010 à 20:33 +0900, Namhyung Kim a écrit :

> What's the difference? I didn't change the semantics of likely clause
> but simply exchange the order of then-else part because IMHO a
> negative expression make it hard to understand, generally.
> I've checked compiled binaries for both cases on x86 and they
> produced exactly same code.
> 


Anyway, re-reading your patch, I just dont get why you use __cond_lock()
two times, but not on this part ?

if (!likely(__cond_lock(&t->sighand->siglock,
			lock_task_sighand(t, &flags))))
	goto ret;

This way, your patch would not be misleading.


Another way to deal with this sparse mess is to use a macro, like we do
for read_trylock() :

#define read_trylock(lock)    __cond_lock(lock, _raw_read_trylock(lock))

-->

#define lock_task_sighand(t, flags) 			\
	__cond_lock(&(t)->sighand->siglock,		\
		    _raw_lock_task_sighand(t, flags))

Re: [PATCH] signal: annotate siglock acquisition

[Namhyung Kim]

On Sat, Sep 25, 2010 at 21:23, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> #define lock_task_sighand(t, flags)                     \
>        __cond_lock(&(t)->sighand->siglock,             \
>                    _raw_lock_task_sighand(t, flags))
>

Looks great! Will use it in a new patch.
Many thanks.


-- 
Regards,
Namhyung Kim