Re: [PATCH RFC 12/20] ctlv: Add device and operation dispatch

[Jiri Pirko]

Fri, Oct 23, 2026 at 04:26:33PM +0200, jgg@xxxxxxxx wrote:
>On Fri, Oct 23, 2026 at 02:58:33PM +0200, Jiri Pirko wrote:
>> >> Jason did some nice handling in the fwctl stuff for
>> >> device alloc and hand over after device_initialize()
>> >> (I've copied it into a few places recently :)
>> >> 
>> >> I'd like to see something similar here as it made
>> >> it easier to follow lifetimes etc.
>> >
>> >This is really at the wrong layer.. Attestation subsystem is the lead
>> >subsystem, it should be allocating the struct device memory and the
>> >driver should see the typical
>> >
>> >  struct attestation_dev *adev;
>> >
>> >  adev = attestation_alloc()
>> >  [..]
>> >  attestation_register(adev)
>> 
>> Er, the driver would call attestation_create() which will be a wrapper
>> of ctvl_create. The reason is inheritance chain and each parent being
>> passed info to and doing its own init()
>
>We really want it to look like the above so we get proper container_of
>stuff.
>
>struct driver_attestion_dev {
>   struct attestaion_dev adev;
>}
>
>driver_adev = attestaion_alloc(..)
>
>Don't want "create" to allocate the memory, it is upside down. The
>fixed memory ctvl needs should be buried in the attestaion_Dev
>
>struct attestation_dev {
>   struct device dev;    // sysfs dev inside class attestation
>   struct ctlv_XX ctlv;  // ctlv instance
>   struct cdev cdev;
>
>   const struct attestation_ops *ops;
>}
>
>That's the essential skeleton of what a subsystem should look like in
>linux, ctvl shouldn't co-opt it, it should embed like cdev does.
>
>Look at fwctl for reference

Yeah, I understand the typical flow. I just wanted to benefit from the
fact the ctlv has the inheritance chain to reuse it for init/fini flows
alongside the private structures allocations. It saves a lot of
boilerplate code and hides it inside CTLV. I admit, it makes things a
bit obfuscated for end user (driver/subsystem). I'm working on
conversion to what you suggest.