Re: [PATCH] arm64/mm: Ensure adequate HUGE_MAX_HSTATE

[Ryan Roberts]

On 02/12/2024 06:44, Anshuman Khandual wrote:
> This asserts that HUGE_MAX_HSTATE is sufficient enough preventing potential
> hugetlb_max_hstate runtime overflow in hugetlb_add_hstate() thus triggering
> a BUG_ON() there after.
> 
> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
> Cc: Will Deacon <will@xxxxxxxxxx>
> Cc: Ard Biesheuvel <ardb@xxxxxxxxxx>
> Cc: Ryan Roberts <ryan.roberts@xxxxxxx>
> Cc: Mark Rutland <mark.rutland@xxxxxxx>
> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
> ---
> This patch applies on v6.13-rc1
> 
>  arch/arm64/mm/hugetlbpage.c | 12 ++++++++++++
>  1 file changed, 12 insertions(+)
> 
> diff --git a/arch/arm64/mm/hugetlbpage.c b/arch/arm64/mm/hugetlbpage.c
> index 3215adf48a1b..98a2a0e64e25 100644
> --- a/arch/arm64/mm/hugetlbpage.c
> +++ b/arch/arm64/mm/hugetlbpage.c
> @@ -519,6 +519,18 @@ pte_t huge_ptep_clear_flush(struct vm_area_struct *vma,
>  
>  static int __init hugetlbpage_init(void)
>  {
> +	/*
> +	 * HugeTLB pages are supported on maximum four page table
> +	 * levels (PUD, CONT PMD, PMD, CONT PTE) for a given base
> +	 * page size, corresponding to hugetlb_add_hstate() calls
> +	 * here.
> +	 *
> +	 * HUGE_MAX_HSTATE should at least match maximum supported
> +	 * HugeTLB page sizes on the platform. Any new addition to
> +	 * supported HugeTLB page sizes will also require changing
> +	 * HUGE_MAX_HSTATE as well.
> +	 */
> +	BUILD_BUG_ON(HUGE_MAX_HSTATE < 4);

I guess we only technically need 3 if pud_sect_supported() == false ? Although
for simplicity (and to reduce my rebasing effort for boot-time page size :) ),
agree we should always just compare to 4.

Reviewed-by: Ryan Roberts <ryan.roberts@xxxxxxx>

>  	if (pud_sect_supported())
>  		hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT);
>