Re: [PATCH v4 6/9] dax: plumb hotplug online_type through dax

["David Hildenbrand (Arm)" <david@kernel.org>]

On 6/5/26 23:54, Gregory Price wrote:
> On Fri, Jun 05, 2026 at 09:31:43PM +0000, sashiko-bot@kernel.org wrote:
>>
>>> diff --git a/drivers/dax/kmem.c b/drivers/dax/kmem.c
>>> index 592171ec10f49..41ccb618a1464 100644
>>> --- a/drivers/dax/kmem.c
>>> +++ b/drivers/dax/kmem.c
>>> @@ -172,8 +172,9 @@ static int dev_dax_kmem_probe(struct dev_dax *dev_dax)
>>>  		 * Ensure that future kexec'd kernels will not treat
>>>  		 * this as RAM automatically.
>>>  		 */
>>> -		rc = add_memory_driver_managed(data->mgid, range.start,
>>> -				range_len(&range), kmem_name, mhp_flags);
>>> +		rc = __add_memory_driver_managed(data->mgid, range.start,
>>> +				range_len(&range), kmem_name, mhp_flags,
>>> +				dev_dax->online_type);
>>
>> [Severity: High]
>> Does capturing the auto-online policy at device creation time break dynamic
>> sysfs policy changes?
>>
> 
> @David - we talked about this ~6+ mo ago at this point, i originally
> proposed something like MMOP_SYSTEM_DEFAULT
> 
> https://lore.kernel.org/linux-mm/20260114085201.3222597-3-gourry@gourry.net/
> 
> https://lore.kernel.org/linux-mm/c4ed9675-269c-4764-86d5-87f4f83fc74d@kernel.org/
> 
> '''
>    I don't like having fake options as part of this interface.
> 
>    Why can't we let selected users use mhp_get_default_online_type() 
>    instead? Like add_memory_resource(). We can export that function.
> '''
> 
> I guess the argument here is that the value can't be cached?
> 
> But we can't really enforce that.
> 
> This is a behavioral change we have to agree on (either we change it or
> carry something like MMOP_SYSTEM_DEFAULT for this).

I don't think we'd need a generic MMOP_* for that purpose (the core cannot
really do a lot with that AFAIKS, and the scenario "unspecified" doesn't really
occur in memory hotplug core).

I guess you could have some indication whether the user already configured it
manually (bool online_type_specified), and handle that in the handful of places.

if (dev_dax->online_type_specified)
	online_type = dev_dax->online_type;
else
	online_type = mhp_get_default_online_type();

Of course, you could also just call the old interface when adding without a
preference.

A negative online_type could also be used as "unspecified" in your code (MMOP_
starts at 0).

But the question really is which semantics we would want? is that exposed to
user-space somehow, such that the user could query it ("default")? or how does
this get set, and how does it later get used while it could have changed
system-wide?

-- 
Cheers,

David