Re: [PATCH kernel 6/6] crypto/ccp: Implement SEV-TIO PCIe IDE (phase1)

[Alexey Kardashevskiy]

On 21/11/25 08:28, dan.j.williams@xxxxxxxxx wrote:
> Alexey Kardashevskiy wrote:
> [..]
> > > > +/*to_pci_tsm_pf0((pdev)->tsm)*/
> > >
> > > Left over of something?
> >
> > Actually not, to_pci_tsm_pf0() is a static helper in drivers/pci/tsm.c
> > and pdev_to_tsm_pf0() (below) is the same thing defined for
> > drivers/crypto/ccp/sev-dev-tsm.c and I wonder if to_pci_tsm_pf0() is
> > better be exported. pdev_to_tsm_pf0() does not need all the checks as
> > if we are that far past the initial setup, we can skip on some checks
> > which to_pci_tsm_pf0() performs so I have not exported
> > to_pci_tsm_pf0() but left the comment. Thanks,
>
> Why does the low-level TSM driver need to_pci_tsm_pf0() when it
> allocated the container for @tsm in the first place?


If the question "can I skip pci_tsm_pf0 and cast straight to tsm_dsm_tio" - yes I can and probably will, so so many leftovers are still there :)

If the question why I need pf0 in the TSM driver - for things like tdi_bind/unbind which take VFs pdev and I'll need PF0 for DOE.

Thanks,


> For example, samples/devsec/ does this:
>
> static void devsec_link_tsm_pci_remove(struct pci_tsm *tsm)
> {
>          struct pci_dev *pdev = tsm->pdev;
>
>          dev_dbg(pci_tsm_host(pdev), "%s\n", pci_name(pdev));
>
>          if (is_pci_tsm_pf0(pdev)) {
>                  struct devsec_tsm_pf0 *devsec_tsm = to_devsec_tsm_pf0(tsm);
>
>                  pci_tsm_pf0_destructor(&devsec_tsm->pci);
>                  kfree(devsec_tsm);
>          } else {
>                  struct devsec_tsm_fn *devsec_tsm = to_devsec_tsm_fn(tsm);
>
>                  kfree(devsec_tsm);
>          }
> }
>
> ...where that to_devsec_tsm_pf0() is:
>
> static struct devsec_tsm_pf0 *to_devsec_tsm_pf0(struct pci_tsm *tsm)
> {
>          return container_of(tsm, struct devsec_tsm_pf0, pci.base_tsm);
> }

--
Alexey