Re: commit de752774f38bb causes fatal error on boot

[Bert Karwatzki]

Am Dienstag, dem 10.09.2024 um 15:56 +0200 schrieb Peter Zijlstra:
> On Tue, Sep 10, 2024 at 01:11:09PM +0200, Bert Karwatzki wrote:
> > When booting linux-next-20240910 on my MSI alpha 15 Laptop running debian sid (amd64),
> > I get dropped to a shell and get the folllowing error in dmesg. I bisected this to
> > commit de752774f38bb ("jump_label: Fix static_key_slow_dec() yet again").
>
> I've just replaced that commit with the below -- which should be in
> tomorrows tree:
>
> ---
> commit 1d7f856c2ca449f04a22d876e36b464b7a9d28b6
> Author: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Date:   Mon Sep 9 12:50:09 2024 +0200
>
>     jump_label: Fix static_key_slow_dec() yet again
>
>     While commit 83ab38ef0a0b ("jump_label: Fix concurrency issues in
>     static_key_slow_dec()") fixed one problem, it created yet another,
>     notably the following is now possible:
>
>       slow_dec
>         if (try_dec) // dec_not_one-ish, false
>         // enabled == 1
>                                     slow_inc
>                                       if (inc_not_disabled) // inc_not_zero-ish
>                                       // enabled == 2
>                                         return
>
>         guard((mutex)(&jump_label_mutex);
>         if (atomic_cmpxchg(1,0)==1) // false, we're 2
>
>                                     slow_dec
>                                       if (try-dec) // dec_not_one, true
>                                       // enabled == 1
>                                         return
>         else
>           try_dec() // dec_not_one, false
>           WARN
>
>     Use dec_and_test instead of cmpxchg(), like it was prior to
>     83ab38ef0a0b. Add a few WARNs for the paranoid.
>
>     Fixes: 83ab38ef0a0b ("jump_label: Fix concurrency issues in static_key_slow_dec()")
>     Reported-by: "Darrick J. Wong" <djwong@xxxxxxxxxx>
>     Tested-by: Klara Modin <klarasmodin@xxxxxxxxx>
>     Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
>
> diff --git a/kernel/jump_label.c b/kernel/jump_label.c
> index 6dc76b590703..93a822d3c468 100644
> --- a/kernel/jump_label.c
> +++ b/kernel/jump_label.c
> @@ -168,7 +168,7 @@ bool static_key_slow_inc_cpuslocked(struct static_key *key)
>  		jump_label_update(key);
>  		/*
>  		 * Ensure that when static_key_fast_inc_not_disabled() or
> -		 * static_key_slow_try_dec() observe the positive value,
> +		 * static_key_dec_not_one() observe the positive value,
>  		 * they must also observe all the text changes.
>  		 */
>  		atomic_set_release(&key->enabled, 1);
> @@ -250,7 +250,7 @@ void static_key_disable(struct static_key *key)
>  }
>  EXPORT_SYMBOL_GPL(static_key_disable);
>
> -static bool static_key_slow_try_dec(struct static_key *key)
> +static bool static_key_dec_not_one(struct static_key *key)
>  {
>  	int v;
>
> @@ -274,6 +274,14 @@ static bool static_key_slow_try_dec(struct static_key *key)
>  		 * enabled. This suggests an ordering problem on the user side.
>  		 */
>  		WARN_ON_ONCE(v < 0);
> +
> +		/*
> +		 * Warn about underflow, and lie about success in an attempt to
> +		 * not make things worse.
> +		 */
> +		if (WARN_ON_ONCE(v == 0))
> +			return true;
> +
>  		if (v <= 1)
>  			return false;
>  	} while (!likely(atomic_try_cmpxchg(&key->enabled, &v, v - 1)));
> @@ -284,15 +292,27 @@ static bool static_key_slow_try_dec(struct static_key *key)
>  static void __static_key_slow_dec_cpuslocked(struct static_key *key)
>  {
>  	lockdep_assert_cpus_held();
> +	int val;
>
> -	if (static_key_slow_try_dec(key))
> +	if (static_key_dec_not_one(key))
>  		return;
>
>  	guard(mutex)(&jump_label_mutex);
> -	if (atomic_cmpxchg(&key->enabled, 1, 0) == 1)
> +	val = atomic_read(&key->enabled);
> +	/*
> +	 * It should be impossible to observe -1 with jump_label_mutex held,
> +	 * see static_key_slow_inc_cpuslocked().
> +	 */
> +	if (WARN_ON_ONCE(val == -1))
> +		return;
> +	/*
> +	 * Cannot already be 0, something went sideways.
> +	 */
> +	if (WARN_ON_ONCE(val == 0))
> +		return;
> +
> +	if (atomic_dec_and_test(&key->enabled))
>  		jump_label_update(key);
> -	else
> -		WARN_ON_ONCE(!static_key_slow_try_dec(key));
>  }
>
>  static void __static_key_slow_dec(struct static_key *key)
> @@ -329,7 +349,7 @@ void __static_key_slow_dec_deferred(struct static_key *key,
>  {
>  	STATIC_KEY_CHECK_USE(key);
>
> -	if (static_key_slow_try_dec(key))
> +	if (static_key_dec_not_one(key))
>  		return;
>
>  	schedule_delayed_work(work, timeout);

Just tested the new version, no error here.

Bert karwatzki