Re: [RFC PATCH v2 0/3] cxl: Provide explicit RAM region creation for Type-2 providers

[Lucero Palau, Alejandro]

On 14/10/2026 18:16, Dave Jiang wrote:
> On 10/14/26 9:35 AM, Lucero Palau, Alejandro wrote:
> > On 14/10/2026 17:23, Dave Jiang wrote:
> > > On 10/14/26 8:34 AM, Lucero Palau, Alejandro wrote:
> > > > On 07/10/2026 10:05, Richard Cheng wrote:
> > > > > A Type-2 provider uses devm_cxl_probe_mem() to register its memdev
> > > > > and obtain an HPA range. This requires an existing committed region,
> > > > > so registration fails when firmware has not configured one.
> > > > >
> > > > > Firmware may provide a suitable CFMWS window without programming a
> > > > > region.
> > > > Hi Richard,
> > > >
> > > >
> > > > FWIW, I unsuccessfully  tried to explain this possibility to "BIOS people". The fact is, nowadays for AMD and I think for Intel as well, if there is no CXL device detected with CXL.mem enabled, the BIOS will neither set HDMs  nor there will be a CFMWS.
> > > Hmmmm....don't they have to reserve CFMWS for hotplug?
> >
> > I have to admit I'm a bit lost with hotplugging and CXL needs, and not real experience with hot plug devices at all, but in my mind hotplugging allows a great flexibility which is counterintuitive if you need to know before hand which devices are going to be hot plugged ..., regarding CXL, how a CFMWS can be created without having that information? Maybe there is something in the specs about this and I did not pay attention.
> I believe BIOS needs to set aside a range of HPA via CFMWS to allow for hot plugged devices.
>
> > What do you think about my other concern below?
> >
> >
> > > DJ
> > >
> > > > > Hot-added devices can have the same requirement. A provider
> > > > > needs its endpoint topology established before deciding when to
> > > > > create a region and how much memory to map.
> > > > Another unsuccessfully discussion was to avoid BIOS setting HDMs at hotplug time as this could race with host dynamic regions creation.
> The BIOS does not touch device when hot plugged I believe. It's all OS native handled.


If this is the kernel expectation and I'm the only one concerned with 
BIOSes respecting it, nothing more to be discussed.


Thanks,

Alejandro.


> > > > I'm commenting these two things here for maybe having a community/kernel discussion with the goal of defining what BIOSes should do and mostly what they should not do if they want to be compatible with kernel expectations.