Re: [PATCH] powerpc/smp: Handle missing ibm,thread-groups in parse_thread_groups()

[Shrikanth Hegde <sshegde@linux.ibm.com>]

Hi Ritesh,

+cc nathan, Kees,

On 11/10/26 7:13 PM, Ritesh Harjani (IBM) wrote:
> Power systems can have ibm,thread-groups properly absent like in one of
> the PowerNV Power9 boxes where there was a warning splat reported on
> 7.4-rc2:
> 
>    WARNING: mm/slub.c:5459 at ___kmalloc_large_node+0x174/0x1a0, CPU#124: swapper/124/1
>    ...
>    Call Trace:
>     __kmalloc_noprof+0x490/0x690
>     _kzalloc_noprof.constprop.0+0x1c/0x7c
>     init_thread_group_cache_map+0xd0/0x2d4
>     smp_prepare_cpus+0x2bc/0x484
>     kernel_init_freeable+0x168/0x3c4
>     kernel_init+0x34/0x1ac
> 
> This happens because today we don't check for whether the properly is
> present or not and directly pass the return value of
> of_property_count_u32_elems(), to the allocator.
> 
> This went unnoticed because earlier kcalloc() caught the overflow due to
> check_mul_overflow() check and returned NULL.
> But since commit 0af298e62e5c ("treewide: refresh kmalloc_obj()
> conversions") switched this call to kzalloc_objs(), size_mul() saturates
> the size to SIZE_MAX instead and we now get a warning at boot.
> 
> Bail out early if the property is missing or empty. Follow the return
> value as documented in kernel doc of parse_thread_groups() function.
> 
> Fixes: 790a1662d3a2 ("powerpc/smp: Parse ibm,thread-groups with multiple properties")
> Reported-by: Madhavan Srinivasan <maddy@linux.ibm.com>
> Signed-off-by: Ritesh Harjani (IBM) <ritesh.list@gmail.com>
> ---
>   arch/powerpc/kernel/smp.c | 4 ++++
>   1 file changed, 4 insertions(+)
> 
> diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> index 7a25eaa59e62..9a0f63e43821 100644
> --- a/arch/powerpc/kernel/smp.c
> +++ b/arch/powerpc/kernel/smp.c
> @@ -824,6 +824,10 @@ static int parse_thread_groups(struct device_node *dn,
>   	int i = 0;
> 
>   	count = of_property_count_u32_elems(dn, "ibm,thread-groups");
> +	if (count < 0)
> +		return count;
> +	if (!count)
> +		return -ENODATA;

This zero count check is not necessary.

Code doesn;t dereference thread_group_array if the count is 0.

...
         while (i < count && property_idx < MAX_THREAD_GROUP_PROPERTIES) {
...
out_free:
         kfree(thread_group_array);
         return ret;

Which should naturally pass to kfree which is expected to handle ZERO_SIZE_PTR.

No?

>   	thread_group_array = kzalloc_objs(u32, count);
>   	if (!thread_group_array)
>   		return -ENOMEM;
> --
> 2.39.5
> 

Nathan reported the same issue on QEMU.The same symptoms.

https://lore.kernel.org/all/20261109-powerpc-smp-handle-count-err-v1-1-e777fde7f71a@kernel.org/