Dan Williams <dan.j.williams@xxxxxxxxx> writes:
> Aneesh Kumar K.V wrote:
>> Dan Williams <dan.j.williams@xxxxxxxxx> writes:
>>
>> ....
>>
>> > +static void pci_tsm_pf0_init(struct pci_dev *pdev)
>> > +{
>> > + bool tee_cap;
>> > +
>> > + tee_cap = pdev->devcap & PCI_EXP_DEVCAP_TEE;
>> > +
>> > + if (!(pdev->ide_cap || tee_cap))
>> > + return;
>> >
>>
>> If we expect to use pci_tsm_pf0_init and is_pci_tsm_pf0() from the
>> guest, can we have the ide_cap and tee_cap check here? Will that be true
>> for all devices assigned to the guest?
>
> I do not expect this path to be taken for a guest device. IDE is not
> relevant for TDIs in the guest and function0 is not a requirement guest
> BDFs. I still need to add this to samples/devsec/, but the expectation
> is that in "TVM mode" the presence of PCI_EXP_DEVCAP_TEE is sufficient
> to route any PCI function to tsm_ops->probe().
We do that because we expose /sys/bus/pci/devices/<x>/tsm/connect to
guest and we have
static int pci_tsm_connect(struct pci_dev *pdev)
{
struct pci_tsm_pf0 *tsm = to_pci_tsm_pf0(pdev->tsm);
int rc;
....
if (tvm_mode())
rc = __driver_idle_connect(pdev);
else
rc = tsm_ops->connect(pdev);
-aneesh
