[RFC PATCH 2/2] firmware: arm_scmi: Add bus support for autoloading

Cristian Marussi cristian.marussi at arm.com
Mon Jun 8 13:53:00 PDT 2026


On Mon, Jun 08, 2026 at 07:06:42PM +0200, Daniel Lezcano wrote:
> 
> Hi Cristian,
> 
> thanks for your answer
> 
> On 6/8/26 18:51, Cristian Marussi wrote:
> > On Mon, Jun 08, 2026 at 04:51:03PM +0200, Daniel Lezcano wrote:
> > > On Mon, Feb 03, 2025 at 10:01:54AM +0000, Cristian Marussi wrote:
> > > > Emit proper MODALIAS uevents when SCMI devices are created and make sure
> > > > all the standard protocol devices are requested when the bus is
> > > > initialized.
> > > > 
> > > > Signed-off-by: Cristian Marussi <cristian.marussi at arm.com>
> > > > ---
> > > 
> > > Hi Cristian,

Hi,

> > > 
> > 
> > Hi Daniel,
> > 
> > nice to hear from you in the SCMI land :P
> > 
> > > what is the status of this patch ?
> > 
> > ....I'd say forgotten/abandoned...I worked on that a bit when I realized
> > only part of the stack was autoloaded...then it did NOT received much
> > love and then I forgot it....buried by other prios....
> > 
> > Indeed I have still the local branch...but after a quick check I think
> > is the same as the RFC posted.
> > 
> > bbd14b0f7733 firmware: arm_scmi: Add bus support for autoloading
> > 8d982393b505 firmware: arm_scmi: Generate aliases for SCMI modules
> > 383c127faa97 (scmi_vendors_autoload_V2) firmware: arm_scmi: Add aliases to transport modules
> > 00caa894bce2 firmware: arm_scmi: Add module aliases to i.MX vendor protocols
> > d900620c46bb firmware: arm_scmi: Support vendor protocol modules autoloading
> > 4fe57bbeb6dc firmware: arm_scmi: Allow transport properties for multiple instances
> > ad236e5a7f01 Linux 6.13-rc1
> > 
> > ...where scmi_vendors_autoload_V2 is merged already...
> 
> Actually, I'm puzzled.
> 
> On our platform, until 7.1-rc1 we had to add a modprobe.d script to load the
> scmi_cpufreq driver because autoload was not suppported.
> 
> Now (7.1-rc1) it seems to be automatically loaded.
>

mmm...I am not sure....but I have a memory to have seen this behaviour
with cpufreq...on some more full-fledged distro (not the usual bare
minimum deboostrapped thing...)...never fully investigated though...

> I imagined the autoload module has been added between 7.0 and 7.1, but the
> series you are mentioning is from 6.13.

Not sure when effectively merged BUT definitely NOT 7.0/7.1...

> 
> What I am missing ?

Any chance that on a more complete kernel/distro some symbol dependencies
in modules.dep kicks in, unknowingly, that triggers the load of
scmi-coufreq ?

Not sure if it make any sense...since I miss anyway where the
SCMI-cpufreq <---> cpufreq-driver association is baked in with the current
module device tables...

Not so much of an help here...sorry.

Anyway, I may respin this in the future if there is some interest...unless
someone precedes me :D

Thanks
Cristian




More information about the linux-arm-kernel mailing list