On 08-05-2025 01:19, Rodrigo Vivi wrote:
+static int xe_pcode_fwctl_uctx_open(struct fwctl_uctx *uctx)
+{
+ struct xe_pcode_fwctl_dev *fwctl_dev =
+ container_of(uctx->fwctl, struct xe_pcode_fwctl_dev, fwctl);
+ struct xe_device *xe = fwctl_dev->xe;
+
+ xe_pm_runtime_get(xe);
Shouldn't this be in the RPC function? Why keep the device awake as
long as a the FD is open?
In general I prefer the runtime on the outer bounds to avoid funny deadlock
cases. But right, in this case here it could be inside the xe_pcode calls
itself, that is when the mmio/mailboxes accesses will really happen.
This I will handle in separate patch. For now will do runtime on rpc
function.
Regards,
Badal
