On Mon, Mar 6, 2023 at 11:49 AM Andy Shevchenko [off-list ref] wrote:

On Thu, Mar 02, 2023 at 06:06:06PM +0100, Benjamin Tissoires wrote:

quoted

On Mar 01 2023, Andy Shevchenko wrote:

quoted

On Tue, Feb 28, 2023 at 01:05:54PM -0600, Daniel Kaehn wrote:

...

quoted

quoted

[1]: https://stackoverflow.com/a/60855157/2511795

Thanks Andy for your help here, and thanks for that link.

I am trying to test Danny's patch as I want to use it for my HID CI,
being an owner of a CP2112 device myself.

The current setup is using out of the tree patches [2] which are
implementing a platform i2c-hid support and some manual addition of a
I2C-HID device on top of it. This works fine but gets busted every now
and then when the tree sees a change that conflicts with these patches.

So with Danny's series, I thought I could have an SSDT override to
declare that very same device instead of patching my kernel before
testing it.

Of course, it gets tricky because I need to run that under qemu.

I am currently stuck at the "sharing the firmware_node from usb with
HID" step and I'd like to know if you could help me.

On my laptop, if I plug the CP2112 (without using a USB hub), I can get:

$> ls -l /sys/bus/hid/devices/0003:10C4:EA90.*
  lrwxrwxrwx. 1 root root 0 Mar  2 17:02 /sys/bus/hid/devices/0003:10C4:EA90.0079 -> ../../../devices/pci0000:00/0000:00:14.0/usb2/2-9/2-9:1.0/0003:10C4:EA90.0079
$> ls -l /sys/bus/usb/devices/2-9*/firmware_node
  lrwxrwxrwx. 1 root root 0 Mar  2 17:03 /sys/bus/usb/devices/2-9:1.0/firmware_node -> ../../../../../LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/device:15/device:16/device:25
  lrwxrwxrwx. 1 root root 0 Mar  2 17:02 /sys/bus/usb/devices/2-9/firmware_node -> ../../../../LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/device:15/device:16/device:25

So AFAIU the USB device is properly assigned a firmware node. My dsdt
also shows the "Device (RHUB)" and I guess everything is fine.

Yes, so far so good.

quoted

However, playing with qemu is not so easy.

I am running qemu with the following arguments (well, almost because I
have a wrapper script on top of it and I also run the compiled kernel
from the current tree):

#> qemu-system-x86_64 -bios /usr/share/edk2/ovmf/OVMF_CODE.fd \
                      -netdev user,id=hostnet0 \
                      -device virtio-net-pci,netdev=hostnet0 \
                      -m 4G \
                      -enable-kvm \
                      -cpu host \
                      -device qemu-xhci -usb \
                      -device 'usb-host,vendorid=0x10c4,productid=0xea90' \
                      -cdrom ~/Downloads/Fedora-Workstation-Live-x86_64-37-1.7.iso

Side question, where can I get those blobs from (EDKII and Fedora Live CD)?
I'm using Debian unstable.

You can install the ovmf package in debian[3], which should have a
similar file.
For the Fedora livecd -> https://getfedora.org/fr/workstation/download/
but any other distribution with a recent enough kernel should show the
same.
  >

quoted

And this is what I get:

#> ls -l /sys/bus/hid/devices/0003:10C4:EA90.*
  lrwxrwxrwx 1 root root 0 Mar  2 16:10 /sys/bus/hid/devices/0003:10C4:EA90.0001 -> ../../../devices/pci0000:00/0000:00:06.0/usb2/2-1/2-1:1.0/0003:10C4:EA90.0001

#> ls -l /sys/bus/usb/devices/2-1*/firmware_node
  ls: cannot access '/sys/bus/usb/devices/2-1*/firmware_node': No such file or directory

Looking at the DSDT, I do not see any reference to the USB hub, so I
wonder if the firmware_node needs to be populated first in the DSDT.

So, where QEMU takes DSDT (ACPI tables in general) from? Can you patch that?
I believe that's the problem in qemu.

That's a good question and it's one I am not sure I have the answer to.
I would have assumed that the DSDT was in the OVMF firmware, but given
that we can arbitrarily add command line arguments, I believe it
probably just provides a baseline and then we are screwed. The OVMF bios
is compiled only once, so I doubt there is any mechanism to
enable/disable a component in the DSDT, or make it dynamically
generated.

quoted

Also note that if I plug the CP2112 over a docking station, I lose the
firmware_node sysfs entries on the host too.

This seems like a lack of firmware node propagating in the USB hub code in
the Linux kernel.

That would make a lot of sense.

FWIW, in the VM I see a firmware node on the pci controller itself:
#> ls -l /sys/devices/pci0000\:00/0000\:00\:06.0/firmware_node
   lrwxrwxrwx 1 root root 0 Mar  6 12:24 /sys/devices/pci0000:00/0000:00:06.0/firmware_node -> ../../LNXSYSTM:00/LNXSYBUS:00/PNP0A03:00/device:07

And one the host, through a USB hub:

#> ls -l /sys/bus/hid/devices/0003:10C4:EA90.*
   lrwxrwxrwx. 1 root root 0 Mar  6 13:26 /sys/bus/hid/devices/0003:10C4:EA90.007C -> ../../../devices/pci0000:00/0000:00:14.0/usb2/2-8/2-8.2/2-8.2.4/2-8.2.4:1.0/0003:10C4:EA90.007C
#> ls -l /sys/bus/usb/devices/2-8*/firmware_node
   lrwxrwxrwx. 1 root root 0 Mar  2 16:53 /sys/bus/usb/devices/2-8:1.0/firmware_node -> ../../../../../LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/device:15/device:16/device:1e
   lrwxrwxrwx. 1 root root 0 Mar  2 16:53 /sys/bus/usb/devices/2-8/firmware_node -> ../../../../LNXSYSTM:00/LNXSYBUS:00/PNP0A08:00/device:15/device:16/device:1e

Note that the firmware node propagation stopped at 2-8, and 2.8.2 is not
having a firmware node.

quoted

Do you think it would be achievable to emulate that over qemu and use a
mainline kernel without patches?

As long as qemu provides correct DSDT it should work I assume.

Just to be sure I understand, for this to work, we need the DSDT to
export a "Device(RHUB)"?
Or if we fix the USB fw_node propagation, could we just overwrite
"\_SB_.PCI0.S30_"?  "\_SB_.PCI0.S30_" is the name the ACPI is giving to
the USB port in my VM case AFAIU.

Cheers,
Benjamin

quoted

[2] https://gitlab.freedesktop.org/bentiss/gitlab-kernel-ci/-/tree/master/VM

[3] https://packages.debian.org/buster/all/ovmf/filelist