HDMI passthrough

[Rivanfirdausa]

I am having trouble with HDMI passthrough. Is it actually possible to access HDMI through the Bao hypervisor on the Toradex Verdin IMX8MPlus Dahlia carrier board? Have you tried it before? what is needed and what needs to be changed?

[DavidMCerdeira]

Hi
We've never tested this specifically on this board, but I don't think there's any fundamental limitation.
It might be necessary to allow access to the firmware for some hardware to work correctly: bao-project/bao-demos@4af5306#diff-1a6719032d70a43aa0bb374c76c7c692e097df4dbccb973ebc5560d7028b1167

Aside from this, as long as all the necessary devices (incl. interrupts are assigned) to the VM and you are using place_phys to force VM to use physical addresses, all should work correctly.

[DavidMCerdeira]

Yes I'd say this is necessary

[josecm]

@Rivanfirdausa Yes. This is a cross-cutting issue for all hypervisors that provide device passthrough. Some drivers assume access to other platform-level controls to configure the device which we might not want to allow the guest to access. In this case the access takes the form of SCMI calls to firmware. At the moment the workaround is to selectively allow certain all SMC calls to firmware (for example, see this patch to workaround the same issue on NXP's S32G).

We have been aware of this issue for quite some time and have plans to add support for access control and filtering of SCMI calls before passing them down to firmware (as is done for PSCI). This will probably be addressed in Q3 2026.

[Rivanfirdausa]

Does this patch also apply to GPUs and VPUs? Or is the handling different? The previous HDMI issue has been resolved, and I want to access the GPU that requires Galcore.

[josecm]

I haven’t personally assigned the GPU/VPU to a VM, but I expect the situation to be essentially the same. These blocks are “just” peripherals from the hypervisor’s point of view: they expose MMIO register ranges and raise interrupts, so in principle passthrough is a matter of mapping the relevant MMIO regions and routing the IRQs to the guest.

That said, as with other peripherals on i.MX8*, bring-up may still depend on platform control interfaces (e.g., clocks/resets/power domains/firmware services), so MMIO+IRQs may not be sufficient if the driver relies on firmware calls that are currently blocked/need allowlisting.

I’d recommend trying it and checking the guest dmesg during driver probe and adding INFO prints in Bao to see if the guest issues additional SMCs. I’d be very interested to hear what you find.

[Rivanfirdausa]

Okay, I'll try.