Re: [PATCH v5 3/7] cxl/region: Skip decoder reset on detach for autodiscovered regions

[Alejandro Lucero Palau <alucerop@amd.com>]

On 1/27/26 23:37, dan.j.williams@intel.com wrote:
> Koralahalli Channabasappa, Smita wrote:
> [..]
>> I’m re reading Dan’s note here:
>> https://lore.kernel.org/all/6930dacd6510f_198110020@dwillia2-mobl4.notmuch/
>>
>> Specifically this part:
>> "If the administrator actually wants to destroy and reclaim that
>> physical address space then they need to forcefully de-commit that
>> auto-assembled region via the @commit sysfs attribute. So that means
>> commit_store() needs to clear CXL_REGION_F_AUTO to get the decoder reset
>> to happen."
>>
>> Today the sysfs commit=0 path inside commit_store() resets decoders
>> without the AUTO check whereas the detach path now skips the reset when
>> CXL_REGION_F_AUTO is set.
>>
>> I think the same rationale should apply to the sysfs de-commit path as
>> well? I’m trying to understand the implications of not guarding the
>> reset with AUTO in commit_store().
> Linux tends to give the administrator the ability to know better than the
> kernel. So if the root forcefully decommits the region, root gets to
> keep the pieces.


I have been trying to figure out how to preserve the decoders for Type2 
auto discover regions since, I think, this was demanded after v22 sent 
upstream. This patch/change is what I was looking for, and although I 
did implement it in another way requiring "consensus", this one seems 
good enough and already discussed and approved, so all good.


However, I think it would be also interesting to give the Type2 driver 
the option of resetting decoders as well, what I have been using for v22 
and successfully tested. But this change will preclude that other 
possibility, so, what about an option for clearing CXL_REGION_F_AUTO by 
Type2 drivers? If you want this only to be done by admin/root, I guess a 
module param would do it.