RE: [PATCH] mm/hugetlb: use kvmalloc_array for hugetlb_fault_mutex_table

[Srivastava, Dheeraj Kumar]

AMD General

Hi,

I will send v2.

Thanks
Dheeraj

-----Original Message-----
From: Muchun Song <muchun.song@xxxxxxxxx>
Sent: Tuesday, October 6, 2026 7:41 PM
To: Srivastava, Dheeraj Kumar <DheerajKumar.Srivastava@xxxxxxx>
Cc: osalvador@xxxxxxx; david@xxxxxxxxxx; akpm@xxxxxxxxxxxxxxxxxxxx; davidlohr@xxxxxx; linux-mm@xxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Hegde, Vasant <Vasant.Hegde@xxxxxxx>; Rao, Bharata Bhasker <bharata@xxxxxxx>
Subject: Re: [PATCH] mm/hugetlb: use kvmalloc_array for hugetlb_fault_mutex_table

[You don't often get email from muchun.song@xxxxxxxxx. Learn why this is important at https://aka.ms/LearnAboutSenderIdentification ]

> On Oct 6, 2026, at 16:02, Dheeraj Kumar Srivastava <dheerajkumar.srivastava@xxxxxxx> wrote:
>
> On systems/VMs with a large number of CPUs (e.g. 4096), hugetlb_init()
> computes num_fault_mutexes = roundup_pow_of_two(8 *
> num_possible_cpus()) = 32768.  The subsequent kmalloc for the mutex
> table can exceed MAX_PAGE_ORDER when struct mutex is enlarged by debug
> options like CONFIG_DEBUG_MUTEXES and CONFIG_DEBUG_LOCK_ALLOC, resulting in:
>
>  WARNING: at __alloc_frozen_pages_noprof (order > MAX_PAGE_ORDER)
> kernel BUG at mm/hugetlb.c (BUG_ON(!hugetlb_fault_mutex_table))
>  Kernel panic - not syncing: Fatal exception
>
> Switch to kvmalloc_array() so the allocation falls back to vmalloc
> when the contiguous physical allocation is too large.  The table is
> only accessed by index, so virtual contiguity is sufficient.
>
> Fixes: 8382d914ebf7 ("mm, hugetlb: improve page-fault scalability")
> Signed-off-by: Dheeraj Kumar Srivastava
> <dheerajkumar.srivastava@xxxxxxx>
> ---
> mm/hugetlb.c | 3 ++-
> 1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c index
> cea25773a6c9..f1b982595151 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -4177,7 +4177,8 @@ static int __init hugetlb_init(void)
>       num_fault_mutexes = 1;
> #endif
>       hugetlb_fault_mutex_table =
> -             kmalloc_objs(struct mutex, num_fault_mutexes);
> +             kvmalloc_array(num_fault_mutexes,
> +                            sizeof(*hugetlb_fault_mutex_table),
> + GFP_KERNEL);

Why not use kvmalloc_objs instead? A little simple.

Thanks

>       BUG_ON(!hugetlb_fault_mutex_table);
>
>       for (i = 0; i < num_fault_mutexes; i++)
> --
> 2.25.1
>