Re: [peterz-queue:sched/core 6/9] kernel/sched/topology.c:2705:25: sparse: sparse: incorrect type in assignment (different address spaces)

[Andrea Righi <arighi@nvidia.com>]

On Thu, May 14, 2026 at 09:13:09PM +0800, kernel test robot wrote:
> tree:   https://git.kernel.org/pub/scm/linux/kernel/git/peterz/queue.git sched/core
> head:   238b6f2b3a4948ca7048025a15874b2c253a34bf
> commit: c3d608e99c540bb4a68f31a1e0dffa7af2827216 [6/9] sched/fair: Attach sched_domain_shared to sd_asym_cpucapacity
> config: nios2-randconfig-r123-20260514 (https://download.01.org/0day-ci/archive/20260514/202605142117.gyxGm7SY-lkp@intel.com/config)
> compiler: nios2-linux-gcc (GCC) 8.5.0
> sparse: v0.6.5-rc1
> reproduce (this is a W=1 build): (https://download.01.org/0day-ci/archive/20260514/202605142117.gyxGm7SY-lkp@intel.com/reproduce)
> 

...

>    kernel/sched/build_utility.c: note: in included file:
>    kernel/sched/sched.h:2379:25: sparse: sparse: incompatible types in comparison expression (different address spaces):
>    kernel/sched/sched.h:2379:25: sparse:    struct task_struct [noderef] __rcu *
>    kernel/sched/sched.h:2379:25: sparse:    struct task_struct *
> 
> vim +2705 kernel/sched/topology.c
> 
>   2679	
>   2680	/*
>   2681	 * For asymmetric CPU capacity, attach sched_domain_shared on the innermost
>   2682	 * SD_ASYM_CPUCAPACITY_FULL ancestor of @cpu's base domain when that ancestor is
>   2683	 * not an overlapping NUMA-built domain (then LLC should claim shared).
>   2684	 *
>   2685	 * A CPU may lack any FULL ancestor (e.g., exclusive cpuset symmetric island),
>   2686	 * then LLC must claim shared instead.
>   2687	 *
>   2688	 * Note: SD_ASYM_CPUCAPACITY_FULL is only set when all CPU capacity values
>   2689	 * are present in the domain span, so the asym domain we attach to cannot
>   2690	 * degenerate into a single-capacity group. The relevant edge cases are instead
>   2691	 * covered by the caveats above.
>   2692	 *
>   2693	 * Return true if this CPU's asym path claimed sd->shared, false otherwise.
>   2694	 */
>   2695	static bool claim_asym_sched_domain_shared(struct s_data *d, int cpu)
>   2696	{
>   2697		struct sched_domain *sd = *per_cpu_ptr(d->sd, cpu);
>   2698		struct sched_domain *sd_asym;
>   2699	
>   2700		if (!sd)
>   2701			return false;
>   2702	
>   2703		sd_asym = sd;
>   2704		while (sd_asym && !(sd_asym->flags & SD_ASYM_CPUCAPACITY_FULL))
> > 2705			sd_asym = sd_asym->parent;

This looks like a false positive to me: claim_asym_sched_domain_shared() is only
called from build_sched_domains() and we are walking child -> parent during
sched_domain construction, before the tree is published to RCU readers, so the
code holds exclusive ownership at that point, therefore plain dereference seems
correct.

-Andrea

>   2706	
>   2707		if (!sd_asym || (sd_asym->flags & SD_NUMA))
>   2708			return false;
>   2709	
>   2710		init_sched_domain_shared(d, sd_asym);
>   2711		return true;
>   2712	}
>   2713	
> 
> --
> 0-DAY CI Kernel Test Service
> https://github.com/intel/lkp-tests/wiki