On 27/8/25 09:58, dan.j.williams@xxxxxxxxx wrote:
Alexey Kardashevskiy wrote:
[..trim reply..]
+int pci_tsm_doe_transfer(struct pci_dev *pdev, enum pci_doe_proto type,
+ const void *req, size_t req_sz, void *resp,
+ size_t resp_sz)
This does not belong here yet - no user.
I had been pulling things out without users, but Yilun and Aneesh ask
for them to be included for staging purposes. When this staging branch
goes upstream a user for all exported APIs is a requirement. However,
per your comment below this is worth deleting.
I did not mean to drop it just because of no users but it could stay in a separate patch as this one is quite big (679 insertions) and no-users stuff just adds unnecessary noise.
But if you still want it - "enum pci_doe_proto" should go to pci-doe.h like this https://github.com/AMDESE/linux-kvm/commit/af12dec97ed98a9f365bbbb6925e76c556937d01
Yeah, I will move it over there.
+{
+ struct pci_tsm_pf0 *tsm;
+
+ if (!pdev->tsm || !is_pci_tsm_pf0(pdev))
+ return -ENXIO;
+
+ tsm = to_pci_tsm_pf0(pdev->tsm);
+ if (!tsm->doe_mb)
+ return -ENXIO;
+
+ return pci_doe(tsm->doe_mb, PCI_VENDOR_ID_PCI_SIG, type, req, req_sz,
+ resp, resp_sz);
The wrapper does not seem to be very helpful - the platform driver
(==TSM ==CCP) which is going to call it already knows it is a DSM and
mailboxes are initialized (otherwise the DSM's pci_tsm_ops::probe()
would've failed) so it can just call pci_doe() directly. Thanks,
True, this is a weak helper the TSM driver already knows when and if it
should be using the already exported pci_doe().
right, thanks,
--
Alexey
