Re: [PATCH] rcutorture: Avoid problematic critical section nesting on RT

["Paul E. McKenney" <paulmck@kernel.org>]

On Tue, Aug 17, 2021 at 04:40:18PM +0200, Sebastian Andrzej Siewior wrote:
> From: Scott Wood <swood@redhat.com>
> 
> rcutorture was generating some nesting scenarios that are not
> reasonable.  Constrain the state selection to avoid them.
> 
> Example:
> 
> 1. rcu_read_lock()
> 2. local_irq_disable()
> 3. rcu_read_unlock()
> 4. local_irq_enable()
> 
> If the thread is preempted between steps 1 and 2,
> rcu_read_unlock_special.b.blocked will be set, but it won't be
> acted on in step 3 because IRQs are disabled.  Thus, reporting of the
> quiescent state will be delayed beyond the local_irq_enable().
> 
> For now, these scenarios will continue to be tested on non-PREEMPT_RT
> kernels, until debug checks are added to ensure that they are not
> happening elsewhere.
> 
> Signed-off-by: Scott Wood <swood@redhat.com>
> [valentin.schneider@arm.com: Don't disable BH in atomic context]
> [bigeasy: remove 'preempt_disable(); local_bh_disable(); preempt_enable();
>  local_bh_enable();' from the examples because this works on RT now. ]
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

This looks close to being ready for mainline, actually.

One comment below.

							Thanx, Paul

> ---
> I folded Valentin's bits.
> I removed the unbalanced preempt_disable()/migrate_disable() part from
> the description because it is supported now by the migrate disable
> implementation. I didn't find it explicit in code/ patch except as part
> of local_bh_disable().
> 
> 
>  kernel/rcu/rcutorture.c |   94 ++++++++++++++++++++++++++++++++++++++++--------
>  1 file changed, 80 insertions(+), 14 deletions(-)
> ---
> --- a/kernel/rcu/rcutorture.c
> +++ b/kernel/rcu/rcutorture.c
> @@ -61,10 +61,13 @@ MODULE_AUTHOR("Paul E. McKenney <paulmck
>  #define RCUTORTURE_RDR_RBH	 0x08	/*  ... rcu_read_lock_bh(). */
>  #define RCUTORTURE_RDR_SCHED	 0x10	/*  ... rcu_read_lock_sched(). */
>  #define RCUTORTURE_RDR_RCU	 0x20	/*  ... entering another RCU reader. */
> -#define RCUTORTURE_RDR_NBITS	 6	/* Number of bits defined above. */
> +#define RCUTORTURE_RDR_ATOM_BH	 0x40	/*  ... disabling bh while atomic */
> +#define RCUTORTURE_RDR_ATOM_RBH	 0x80	/*  ... RBH while atomic */
> +#define RCUTORTURE_RDR_NBITS	 8	/* Number of bits defined above. */
>  #define RCUTORTURE_MAX_EXTEND	 \
>  	(RCUTORTURE_RDR_BH | RCUTORTURE_RDR_IRQ | RCUTORTURE_RDR_PREEMPT | \
> -	 RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED)
> +	 RCUTORTURE_RDR_RBH | RCUTORTURE_RDR_SCHED | \
> +	 RCUTORTURE_RDR_ATOM_BH | RCUTORTURE_RDR_ATOM_RBH)
>  #define RCUTORTURE_RDR_MAX_LOOPS 0x7	/* Maximum reader extensions. */
>  					/* Must be power of two minus one. */
>  #define RCUTORTURE_RDR_MAX_SEGS (RCUTORTURE_RDR_MAX_LOOPS + 3)
> @@ -1429,31 +1432,53 @@ static void rcutorture_one_extend(int *r
>  	WARN_ON_ONCE((idxold >> RCUTORTURE_RDR_SHIFT) > 1);
>  	rtrsp->rt_readstate = newstate;
>  
> -	/* First, put new protection in place to avoid critical-section gap. */
> +	/*
> +	 * First, put new protection in place to avoid critical-section gap.
> +	 * Disable preemption around the ATOM disables to ensure that
> +	 * in_atomic() is true.
> +	 */
>  	if (statesnew & RCUTORTURE_RDR_BH)
>  		local_bh_disable();
> +	if (statesnew & RCUTORTURE_RDR_RBH)
> +		rcu_read_lock_bh();
>  	if (statesnew & RCUTORTURE_RDR_IRQ)
>  		local_irq_disable();
>  	if (statesnew & RCUTORTURE_RDR_PREEMPT)
>  		preempt_disable();
> -	if (statesnew & RCUTORTURE_RDR_RBH)
> -		rcu_read_lock_bh();
>  	if (statesnew & RCUTORTURE_RDR_SCHED)
>  		rcu_read_lock_sched();
> +	preempt_disable();
> +	if (statesnew & RCUTORTURE_RDR_ATOM_BH)
> +		local_bh_disable();
> +	if (statesnew & RCUTORTURE_RDR_ATOM_RBH)
> +		rcu_read_lock_bh();
> +	preempt_enable();
>  	if (statesnew & RCUTORTURE_RDR_RCU)
>  		idxnew = cur_ops->readlock() << RCUTORTURE_RDR_SHIFT;
>  
> -	/* Next, remove old protection, irq first due to bh conflict. */
> +	/*
> +	 * Next, remove old protection, in decreasing order of strength
> +	 * to avoid unlock paths that aren't safe in the stronger
> +	 * context.  Disable preemption around the ATOM enables in
> +	 * case the context was only atomic due to IRQ disabling.
> +	 */
> +	preempt_disable();
>  	if (statesold & RCUTORTURE_RDR_IRQ)
>  		local_irq_enable();
> -	if (statesold & RCUTORTURE_RDR_BH)
> +	if (statesold & RCUTORTURE_RDR_ATOM_BH)
>  		local_bh_enable();
> +	if (statesold & RCUTORTURE_RDR_ATOM_RBH)
> +		rcu_read_unlock_bh();
> +	preempt_enable();

The addition of preempt_enable() here prevents rcutorture from covering
an important part of the mainline RCU state space, namely when an RCU
read-side section ends with just local_irq_enable().  This situation
is a challenge for RCU because it must indirectly detect the end of the
critical section.

Would it work for RT if the preempt_enable() and preempt_disable()
were executed only if either RT on the one hand or statesold has the
RCUTORTURE_RDR_ATOM_BH or RCUTORTURE_RDR_ATOM_RBH bit set on the other?

>  	if (statesold & RCUTORTURE_RDR_PREEMPT)
>  		preempt_enable();
> -	if (statesold & RCUTORTURE_RDR_RBH)
> -		rcu_read_unlock_bh();
>  	if (statesold & RCUTORTURE_RDR_SCHED)
>  		rcu_read_unlock_sched();
> +	if (statesold & RCUTORTURE_RDR_BH)
> +		local_bh_enable();
> +	if (statesold & RCUTORTURE_RDR_RBH)
> +		rcu_read_unlock_bh();
> +
>  	if (statesold & RCUTORTURE_RDR_RCU) {
>  		bool lockit = !statesnew && !(torture_random(trsp) & 0xffff);
>  
> @@ -1496,6 +1521,12 @@ rcutorture_extend_mask(int oldmask, stru
>  	int mask = rcutorture_extend_mask_max();
>  	unsigned long randmask1 = torture_random(trsp) >> 8;
>  	unsigned long randmask2 = randmask1 >> 3;
> +	unsigned long preempts = RCUTORTURE_RDR_PREEMPT | RCUTORTURE_RDR_SCHED;
> +	unsigned long preempts_irq = preempts | RCUTORTURE_RDR_IRQ;
> +	unsigned long nonatomic_bhs = RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
> +	unsigned long atomic_bhs = RCUTORTURE_RDR_ATOM_BH |
> +				   RCUTORTURE_RDR_ATOM_RBH;
> +	unsigned long tmp;
>  
>  	WARN_ON_ONCE(mask >> RCUTORTURE_RDR_SHIFT);
>  	/* Mostly only one bit (need preemption!), sometimes lots of bits. */
> @@ -1503,11 +1534,46 @@ rcutorture_extend_mask(int oldmask, stru
>  		mask = mask & randmask2;
>  	else
>  		mask = mask & (1 << (randmask2 % RCUTORTURE_RDR_NBITS));
> -	/* Can't enable bh w/irq disabled. */
> -	if ((mask & RCUTORTURE_RDR_IRQ) &&
> -	    ((!(mask & RCUTORTURE_RDR_BH) && (oldmask & RCUTORTURE_RDR_BH)) ||
> -	     (!(mask & RCUTORTURE_RDR_RBH) && (oldmask & RCUTORTURE_RDR_RBH))))
> -		mask |= RCUTORTURE_RDR_BH | RCUTORTURE_RDR_RBH;
> +
> +	/*
> +	 * Can't enable bh w/irq disabled.
> +	 */
> +	tmp = atomic_bhs | nonatomic_bhs;
> +	if (mask & RCUTORTURE_RDR_IRQ)
> +		mask |= oldmask & tmp;

This is more straightforward than my original, good!

> +
> +	/*
> +	 * Ideally these sequences would be detected in debug builds
> +	 * (regardless of RT), but until then don't stop testing
> +	 * them on non-RT.
> +	 */
> +	if (IS_ENABLED(CONFIG_PREEMPT_RT)) {
> +		/*
> +		 * Can't disable bh in atomic context if bh was already
> +		 * disabled by another task on the same CPU. Instead of
> +		 * attempting to track this, just avoid disabling bh in atomic
> +		 * context.
> +		 */
> +		mask &= ~atomic_bhs;

At some point, we will need to test disabling bh in atomic context,
correct?  Or am I missing something here?

> +		/*
> +		 * Can't release the outermost rcu lock in an irq disabled
> +		 * section without preemption also being disabled, if irqs
> +		 * had ever been enabled during this RCU critical section
> +		 * (could leak a special flag and delay reporting the qs).
> +		 */
> +		if ((oldmask & RCUTORTURE_RDR_RCU) &&
> +		    (mask & RCUTORTURE_RDR_IRQ) &&
> +		    !(mask & preempts))
> +			mask |= RCUTORTURE_RDR_RCU;
> +
> +		/* Can't modify non-atomic bh in atomic context */
> +		tmp = nonatomic_bhs;
> +		if (oldmask & preempts_irq)
> +			mask &= ~tmp;
> +		if ((oldmask | mask) & preempts_irq)
> +			mask |= oldmask & tmp;
> +	}
> +
>  	return mask ?: RCUTORTURE_RDR_RCU;
>  }
>