Re: [PATCH v2 3/6] locking/local_lock: Introduce local_lock_lockdep_start/end()

[Vlastimil Babka <vbabka@suse.cz>]

On 7/15/25 19:29, Alexei Starovoitov wrote:
> On Tue, Jul 15, 2025 at 08:56:21AM +0200, Vlastimil Babka wrote:
>> > the addresses of the locks are different and they're different
>> > kmalloc buckets, but lockdep cannot understand this without
>> > explicit local_lock_lockdep_start().
>> > The same thing I'm trying to explain in the commit log.
>> 
>> Thanks for the explanation and sorry for being so dense.
>> Maybe lockdep's lock classes can be used here somehow instead of having to
>> teach lockdep completely new tricks, but I don't know enough about those to
>> know for sure.
> 
> I tried that with a separate lock_key for each local_trylock_t
> and it's sort-of kinda works for 16 cpus, but doesn't scale
> when number of cpus is large.
> There is no good way to pick LOCKDEP_BITS.
> 
> It can be made optional on PREEMP_RT only
> and used for kmalloc buckets only that kmalloc_nolock() is using,
> but still feels less clean than
> local_lock_lockdep_start/end()
> since it makes lockdep work harder.
> 
> Better ideas?

I was thinking something like a class for normal context and a different
class for kmalloc_nolock() context (determined by allow_spinning) but it's
quite vague as I don't understand lockdep enough.

> From da2b3bac08950929da105836fbff7e2ea4ecbc0e Mon Sep 17 00:00:00 2001
> From: Alexei Starovoitov <ast@kernel.org>
> Date: Tue, 15 Jul 2025 10:16:42 -0700
> Subject: [PATCH] lockdep
> 
> Signed-off-by: Alexei Starovoitov <ast@kernel.org>
> ---
>  lib/Kconfig.debug |  2 +-
>  mm/slub.c         | 13 +++++++++++++
>  2 files changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/Kconfig.debug b/lib/Kconfig.debug
> index ebe33181b6e6..94c07b84ecd0 100644
> --- a/lib/Kconfig.debug
> +++ b/lib/Kconfig.debug
> @@ -1526,7 +1526,7 @@ config LOCKDEP_BITS
>  	int "Size for MAX_LOCKDEP_ENTRIES (as Nth power of 2)"
>  	depends on LOCKDEP && !LOCKDEP_SMALL
>  	range 10 24
> -	default 15
> +	default 16
>  	help
>  	  Try increasing this value if you hit "BUG: MAX_LOCKDEP_ENTRIES too low!" message.
>  
> diff --git a/mm/slub.c b/mm/slub.c
> index 2f30b85fbf68..2ae6bf3ebcd0 100644
> --- a/mm/slub.c
> +++ b/mm/slub.c
> @@ -395,6 +395,7 @@ struct kmem_cache_cpu {
>  	struct slab *partial;	/* Partially allocated slabs */
>  #endif
>  	local_trylock_t lock;	/* Protects the fields above */
> +	struct lock_class_key lock_key;
>  #ifdef CONFIG_SLUB_STATS
>  	unsigned int stat[NR_SLUB_STAT_ITEMS];
>  #endif
> @@ -3083,6 +3084,8 @@ static void init_kmem_cache_cpus(struct kmem_cache *s)
>  	for_each_possible_cpu(cpu) {
>  		c = per_cpu_ptr(s->cpu_slab, cpu);
>  		local_trylock_init(&c->lock);
> +		lockdep_register_key(&c->lock_key);
> +		lockdep_set_class(&c->lock, &c->lock_key);
>  		c->tid = init_tid(cpu);
>  	}
>  }
> @@ -5953,6 +5956,16 @@ void __kmem_cache_release(struct kmem_cache *s)
>  {
>  	cache_random_seq_destroy(s);
>  #ifndef CONFIG_SLUB_TINY
> +	{
> +		int cpu;
> +
> +		for_each_possible_cpu(cpu) {
> +			struct kmem_cache_cpu *c = per_cpu_ptr(s->cpu_slab,
> +							       cpu);
> +
> +			lockdep_unregister_key(&c->lock_key);
> +		}
> +	}
>  	free_percpu(s->cpu_slab);
>  #endif
>  	free_kmem_cache_nodes(s);