Re: [PATCH v2 09/20] hugetlb_vmemmap: Allow architectures to make HVO enablement boot-time only

[Muchun Song <muchun.song@linux.dev>]

> On Oct 3, 2026, at 08:21, James Houghton <jthoughton@google.com> wrote:
> 
> arm64 needs all CPUs to support hardware updates of the access flag to
> safely update the vmemmap in place, and it will need to refuse to online
> CPUs that lack it, but only if HVO may be used. This only works if HVO
> cannot be dynamically enabled, otherwise incompatible CPUs may have
> already been onlined at enable time.
> 
> Add CONFIG_ARCH_WANT_HUGETLB_VMEMMAP_RO_AFTER_INIT for architectures to
> select. When selected, vmemmap_optimize_enabled is __ro_after_init, so
> it can only be set with the hugetlb_free_vmemmap= kernel command line
> parameter, and the sysctl is made read-only.
> 
> Signed-off-by: James Houghton <jthoughton@google.com>
> ---
> Documentation/admin-guide/sysctl/vm.rst |  3 +++
> fs/Kconfig                              |  3 ++-
> mm/Kconfig                              |  9 +++++++++
> mm/hugetlb_vmemmap.c                    | 17 +++++++++++++++--
> 4 files changed, 29 insertions(+), 3 deletions(-)
> 
> diff --git a/Documentation/admin-guide/sysctl/vm.rst b/Documentation/admin-guide/sysctl/vm.rst
> index 5b318d17aa4b..d962d563cfd4 100644
> --- a/Documentation/admin-guide/sysctl/vm.rst
> +++ b/Documentation/admin-guide/sysctl/vm.rst
> @@ -692,6 +692,9 @@ pages.  So, those surplus pages are still optimized until they are no longer
> in use.  You would need to wait for those surplus pages to be released before
> there are no optimized pages in the system.
> 
> +On some architectures, this knob is read-only, and HVO can only be enabled or
> +disabled with the hugetlb_free_vmemmap= kernel command line parameter.
> +
> 
> nr_hugepages_mempolicy
> ======================
> diff --git a/fs/Kconfig b/fs/Kconfig
> index 1454b7fe9641..4222db17ba02 100644
> --- a/fs/Kconfig
> +++ b/fs/Kconfig
> @@ -267,7 +267,8 @@ config HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON
> 	help
> 	  The HugeTLB Vmemmap Optimization (HVO) defaults to off. Say Y here to
> 	  enable HVO by default. It can be disabled via hugetlb_free_vmemmap=off
> -	  (boot command line) or hugetlb_optimize_vmemmap (sysctl).
> +	  (boot command line) or hugetlb_optimize_vmemmap (sysctl, if the
> +	  architecture allows changing it at runtime).
> endif # HUGETLBFS
> 
> config HUGETLB_PAGE
> diff --git a/mm/Kconfig b/mm/Kconfig
> index acefc994d9a8..ad313fddc9da 100644
> --- a/mm/Kconfig
> +++ b/mm/Kconfig
> @@ -475,6 +475,15 @@ config ARCH_WANT_OPTIMIZE_DAX_VMEMMAP
> config ARCH_WANT_OPTIMIZE_HUGETLB_VMEMMAP
> 	bool
> 
> +#
> +# Select this config option from the architecture Kconfig if HugeTLB vmemmap
> +# optimization may only be enabled or disabled on the kernel command line, e.g.

I suggest using the following phrasing to specify precisely what becomes RO.
Something like:

# Select this from the architecture Kconfig if the hugetlb_optimize_vmemmap
# sysctl must be read-only.

> +# because the architecture configures the system at boot based on whether it
> +# is enabled. This makes the hugetlb_optimize_vmemmap sysctl read-only.

I don't think it's necessary to include the last two sentences. This is a public
file, and we only need to describe its purpose. What specifically caused it is
architectural choice, so we don't need to explain it here.

> +#
> +config ARCH_WANT_HUGETLB_VMEMMAP_RO_AFTER_INIT

How about ARCH_HUGETLB_VMEMMAP_OPTIMIZATION_SYSCTL_RO? I'll explain the reason below.

> +	bool
> +
> config HAVE_MEMBLOCK_PHYS_MAP
> 	bool
> 
> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
> index 817ce39d4cf3..de4148a8aabf 100644
> --- a/mm/hugetlb_vmemmap.c
> +++ b/mm/hugetlb_vmemmap.c
> @@ -428,7 +428,20 @@ static int vmemmap_remap_alloc(const struct hstate *h, struct folio *folio,
> 	return ret;
> }
> 
> -static bool vmemmap_optimize_enabled = IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON);
> +/*
> + * Architectures selecting CONFIG_ARCH_WANT_HUGETLB_VMEMMAP_RO_AFTER_INIT rely on
> + * HVO only being enabled or disabled on the kernel command line.
> + */
> +#ifdef CONFIG_ARCH_WANT_HUGETLB_VMEMMAP_RO_AFTER_INIT
> +#define __vmemmap_optimize_enabled_attr	__ro_after_init

If we do not set vmemmap_optimize_enabled to RO, the implementation becomes
much simpler. sysctl is read-only, so it was already impossible to modify
vmemmap_optimize_enabled in the first place. Therefore, even without this
layer of protection, we lose nothing; on the contrary, we gain a simple code
implementation.

> +#define HUGETLB_VMEMMAP_SYSCTL_MODE	0444
> +#else
> +#define __vmemmap_optimize_enabled_attr
> +#define HUGETLB_VMEMMAP_SYSCTL_MODE	0644
> +#endif
> +
> +static bool vmemmap_optimize_enabled __vmemmap_optimize_enabled_attr =
> +	IS_ENABLED(CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP_DEFAULT_ON);
> static int __init hugetlb_vmemmap_optimize_param(char *buf)
> {
> 	return kstrtobool(buf, &vmemmap_optimize_enabled);
> @@ -751,7 +764,7 @@ static const struct ctl_table hugetlb_vmemmap_sysctls[] = {
> 		.procname	= "hugetlb_optimize_vmemmap",
> 		.data		= &vmemmap_optimize_enabled,
> 		.maxlen		= sizeof(vmemmap_optimize_enabled),
> -		.mode		= 0644,
> +		.mode		= HUGETLB_VMEMMAP_SYSCTL_MODE,

So, I think

  .mode = IS_ENABLED(ARCH_HUGETLB_VMEMMAP_OPTIMIZATION_SYSCTL_RO) ? 0444 : 0666,

is simple and enough.

Thanks,
Muchun

> 		.proc_handler	= proc_dobool,
> 	},
> };
> -- 
> 2.56.0.rc1.315.gc6ed9934b7-goog
>