Re: [PATCH 9/9] hw/cxl: Add QMP status query for dynamic-capacity extent release

[Markus Armbruster <armbru@redhat.com>]

Alireza Sanaee <alireza.sanaee@huawei.com> writes:

> On Sat, 28 Mar 2026 07:01:51 +0100
> Markus Armbruster <armbru@redhat.com> wrote:
>
> +CC Ira
>
> Hi Markus,
>
> Please look inline. I believe Ira might be able to help and we can see what he thinks.
>
>> Alireza Sanaee <alireza.sanaee@huawei.com> writes:
>> 
>> > On Thu, 26 Mar 2026 16:15:40 +0100
>> > Markus Armbruster <armbru@redhat.com> wrote:
>> >
>> > Hi Markus,
>> >
>> > This commit adds a test command to check whether an extent (a piece of memory region) has been released or not by the guest.
>> > The process of removal is highly asynchronous, that's why I needed to add a new command to check it.
>> >
>> > I would be happy to drop this patch, as this is extra.  
>> 
>> So your motivation for adding the command is "just" testing.  Correct?
>
> I wrote tests to make sure extents were released, and I used this QMP command to check.

As long as this is the only known use case, we want

    # @unstable: This command is meant for debugging.

in the QAPI schema.

> Now there is background about extents where DCD was developed in Linux by @Ira, CCed [1].
>
> In the DCD extent tracking section of the CXL spec, I see some commands 
> related to getting the extent list, maybe Ira can comment here. To be fair, knowing if an extent
> was released or not is a useful information in general from orchestrator perspective.
>
> [1] https://patchew.org/linux/20250413-dcd-type2-upstream-v9-0-1d4911a0b365@intel.com/

If we believe there are other uses, but we're unsure the command
actually serves them:

    # @unstable: This command is experimental.

>> Can you think of other uses?
>> 
>> In particular, would a management application need this query to monitor
>> removal?
>
> I think management will get a signal on the removal given the schema in Ira's Linux kernel 
> patchset. I do not think this is implemented in QEMU at least (I might be missing something).

I'm getting a feeling of "this is underbaked" :)  But that's for the CXL
maintainers to judge.

>> The description of feature @unstable may need to be adjusted for the
>> answers we work out here.
>
> Absolutely.
>
>> 
>> >
>> > Thanks,
>> > Ali  
>> 
>>