CH9329 doesn't seem to support keyboard or mouse control in some older PC BIOS/UEFI interfaces

[mofosyne]

This is a problem that's affecting multiple KVMs using CH9329 such as JetKVM, nanokvm-usb, glkvm as well as the bare CH9329 to serial cable used in this kvm-serial program... when you try to control the keyboard during UEFI mode of some older PCs like HP G4 prodesk mini PCs (commonly used in homelabs).

I know this is not specific to kvm-serial, but noting here as it be good to hear if anyone had issues with CH9329 as well when in UEFI interface of older PCs.

My best guess is that CH9329 does not know how to speak the simplified HID "Boot Protocol" mode. But I suspect there might be a mode that everyone is forgetting to set to be able to enable such a mode as I find it doubtful that the designer of the CH9329 would have overlooked such requirements.

In case anyone needs it, during my investigations I found these resources quite useful:

• https://github.com/AaLl86/MiniKvm_public/tree/main/Specs
• Tip on reading "CH9329 Chip Serial communication protocol (English).pdf":
  • "work mode X" == "working mode X"
• Used https://ch9329.ayufan.dev/ to inspect the CH9329 configurations

[sjmf]

Following the reverse-engineering work on CH9350 in issue #13, I understand this better enough to have an initial update on this issue too.

So, CH9329 in the kvm-serial implementation is using absolute mouse co-ordinates. Legacy BIOS doesn't support the absolute co-ordinate format. A USB keyboard supporting HID Boot-protocol makes relative mouse reports in the boot protocol fixed format. We could create an operating mode for the software that turns on relative mouse mode. I've just not implemented that, because absolute mouse is so much easier when you can get an x/y coordinate from the mouse position over the video frame.

More crucially, however, the device also needs to advertise bInterfaceSubClass = 0x01 to indicate that it supports boot protocol (see osdev.org). "The boot protocol is much simpler, and follows standard structures for mice and keyboards."

It seems unlikely that CH9329 supports boot protocol. There are no references to the HID protocol spoken in the datasheet, and from your testing it's clear it doesn't automatically. However, it's worth investigating CMD_SET_USB_STRING from the protocol doc (description: "Set the currently used USB string descriptor configuration") to see if this allows configuring the interface class & subclass as a string. It's currently unclear if this could help, and I'm not sure I even own any hardware old enough to test legacy BIOS with.

CH9350L definitely does seem to support it, however, in State 2: both from the datasheet (section 3.3 specifically says: "The mouse is a relative mouse and the keyboard is a standard keyboard under the BIOS protocol") and I've verified this via wire sniffing on real hardware. I've now more or less fully reverse-engineered the protocol for this chip, including State 2. There's a proof-of-concept in issue #13 and I am working on the necessary changes to kvm-serial to integrate it.

I'd like to keep this issue open for a documentation improvement in this repo. We can specifically talk about interfacing with UEFI CSM / Legacy BIOS machines, and when CH9350 is integrated, recommend it.

[sjmf]

Hey @mofosyne - CH9350 support is in the app now 🥳. I was just wondering if you had a chance to get hold of some hardware to test and see if this works for you in legacy BIOS?