Re: [PATCH v4 3/9] mm/memory_hotplug: export mhp_get_default_online_type

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Lazy initialization of `mhp_default_online_type` can race with sysfs updates, silently clobbering the user-specified hotplug policy.
--

commit 7d00f30f087affb00391db7c7b27504af6cf3ac3
Author: Gregory Price <gourry@gourry.net>

mm/memory_hotplug: export mhp_get_default_online_type

Drivers which may pass hotplug policy down to DAX need MMOP_ symbols
and the mhp_get_default_online_type function for hotplug use cases.
This commit exports the mhp_get_default_online_type function to allow
these drivers to build and function when hotplug is disabled.

> diff --git a/mm/memory_hotplug.c b/mm/memory_hotplug.c
> index 6833208cc17cb..4942570540956 100644
> --- a/mm/memory_hotplug.c
> +++ b/mm/memory_hotplug.c
[ ... ]
> @@ -239,6 +239,7 @@ enum mmop mhp_get_default_online_type(void)
>  
>  	return mhp_default_online_type;
>  }
> +EXPORT_SYMBOL_GPL(mhp_get_default_online_type);

[Severity: High]
This isn't a bug introduced by this patch, but exporting this function for
asynchronous driver module initialization paths significantly increases the
probability of a data race with sysfs updates.

Can this lazy initialization clobber a user-specified hotplug policy?

If a driver calls mhp_get_default_online_type() and reads the uninitialized
state just as userspace writes to /sys/devices/system/memory/auto_online_blocks,
could the following sequence occur?

CPU1 (Driver init calling mhp_get_default_online_type):
    if (mhp_default_online_type >= 0) // reads -1

CPU2 (Sysfs write calling auto_online_blocks_store in drivers/base/memory.c):
    mhp_set_default_online_type(online_type); // sets user policy

CPU1 (Continuing in mhp_get_default_online_type):
    if (IS_ENABLED(CONFIG_MHP_DEFAULT_ONLINE_TYPE_OFFLINE))
        mhp_default_online_type = MMOP_OFFLINE; // clobbers user policy

Does this unprotected check-and-set sequence on a global configuration
variable need serialization to prevent the user-specified policy from being
silently lost?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260605211911.2160954-1-gourry@gourry.net?part=3