On Tue, Mar 7, 2023 at 7:17 AM Benjamin Tissoires
[off-list ref] wrote:

On Mar 06 2023, Andy Shevchenko wrote:

quoted

On Mon, Mar 06, 2023 at 01:40:16PM -0600, Daniel Kaehn wrote:

...

quoted

Device (SE9)
{
    Name (_ADR, 0x001D0001) // _ADR: Address
    Device (RHUB)
    {
        Name (_ADR, Zero)
        Device (CP2) // the USB-hid & CP2112 shared node
        {
            Name (_ADR, One)
        }
    }
}

If I'm understanding correctly, this adds the SE9 device as function 1
of PCI device 0x1d,

To be precise this does not add the device. It adds a description of
the companion device in case the real one will appear on the PCI bus
with BDF 00:1d.1.

quoted

then RHUB as the USB controller it provides, and finally, CP2 as the
USB device attached to port 1 of the controller.

With this as the loaded dsdt table, the USB device now has a firmware_node :)
#> cat /sys/bus/usb/devices/3-1:1.0/firmware_node/path
\_SB_.PCI0.SE9_.RHUB.CP2_

After applying my patches, the HID device also references this node:
#> cat /sys/bus/hid/devices/0003:10C4:EA90.0003/firmware_node/path
\_SB_.PCI0.SE9_.RHUB.CP2_

Great! Thanks a lot for that. Turns out that with both of your inputs I
can also do the same, but without the need for OVMF and DSDT patching,
with just an SSDT override.

Ah, interesting.. I tried the SSDT override route initially, but tried
applying it
through EFI variables and through configfs, neither of which worked since
they appeared to be applied after the relevant drivers were already loaded
(at least that was my suspicion). I wonder if loading the overlay through the
initramfs was the key.

Turns out that the override documentation [1] mentions "This option
allows loading of user defined SSDTs from initrd and it is useful when
the system does not support EFI or ..."

FWIW, I am attaching my full DSDT override in case it is valuable:
(on my system, the default USB controller (non-xhc) is at PCI address
1.2, which explains the slight difference). It can be loaded in the same
way you are overriding the full DSDT, but with just that compilation
output:

---
DefinitionBlock ("cp2112.aml", "SSDT", 5, "", "CP2112", 1)
{
  External (_SB_.PCI0, DeviceObj)

  Scope (\_SB_.PCI0)
  {
    Device (USB0)
    {
      Name (_ADR, 0x00010002) // _ADR: Address
      Device (RHUB)
      {
        Name (_ADR, Zero)
        Device (CP21) // the USB-hid & CP2112 shared node
        {
          Name (_ADR, One)
          Device (I2C)
          {
            Name (_ADR, Zero)
            Name (_STA, 0x0F)
          }

          Device (GPIO)
          {
            Name (_ADR, One)
            Name (_STA, 0x0F)
          }
        }
      }
    }
  }

To get this to work -- I assume you had to change the driver to look
for uppercase
"GPIO" and "I2C", or some similar change?

  Scope (\_SB_.PCI0.USB0.RHUB.CP21.I2C)
  {
    Device (TPD0)
    {
      Name (_HID, "RMI40001")
      Name (_CID, "PNP0C50")
      Name (_STA, 0x0F)

      Name (SBFB, ResourceTemplate ()
      {
          I2cSerialBusV2 (0x00c, ControllerInitiated, 100000,
              AddressingMode7Bit, "\\_SB_.PCI0.USB0.RHUB.CP21.I2C",
              0x00, ResourceConsumer,, Exclusive,
              )
      })
      Name (SBFG, ResourceTemplate ()
      {
          GpioInt (Level, ActiveLow, Exclusive, PullDefault, 0x0000,
              "\\_SB_.PCI0.USB0.RHUB.CP21.GPIO", 0x00, ResourceConsumer, ,
              )
              {   // Pin list
                  0x0002
              }
      })
      Method(_CRS, 0x0, NotSerialized)
      {
        Return (ConcatenateResTemplate (SBFB, SBFG))
      }

      Method(_DSM, 0x4, Serialized)
      {
        // DSM UUID
        switch (ToBuffer (Arg0))
        {
          // ACPI DSM UUID for HIDI2C
          case (ToUUID ("3CDFF6F7-4267-4555-AD05-B30A3D8938DE"))
          {
              // DSM Function
              switch (ToInteger (Arg2))
              {
                  // Function 0: Query function, return based on revision
                  case(0)
                  {
                      // DSM Revision
                      switch (ToInteger (Arg1))
                      {
                          // Revision 1: Function 1 supported
                          case (1)
                          {
                              Return (Buffer (One) { 0x03 })
                          }

                          default
                          {
                              // Revision 2+: no functions supported
                              Return (Buffer (One) { 0x00 })
                          }
                      }
                  }

                  // Function 1 : HID Function
                  case(1)
                  {
                      // HID Descriptor Address
                      Return (0x0020)
                  }

                  default
                  {
                      // Functions 2+: not supported
                      Return (Buffer (One) { 0x00 })
                  }
              }
          }

          default
          {
              // No other GUIDs supported
              Return (Buffer (One) { 0x00 })
          }
        }
      }
    }
  }
}
---

This almost works. Almost because the I2C device is correctly created,
but I have an issue with the GpioInt call which is not properly set by
the kernel and which returns -EDEFER. /o\

Ahh, yep, I've had this issue as well -- I suspect the issue you're
having is that the
CP2112 driver initializes the i2c controller before the gpiochip, and
if any i2c devices
on the bus depend on the CP2112's gpio, the probe will never succeed!
I have made
and been testing with a patch to fix this, but since it was midway
through submitting
this series, thought it might be bad practice to "tack on" additional
patches to a patchset
mid-review (since it only causes an issue in some (admittedly fairly
common) use-cases)
so I was going to send it as an individual patch once (if) these were applied.

If you think that would be necessary to include for these to merge,
I'd be happy to append
it to this review. I also have another patch which adds i2c bus
recovery to the driver, but
that seems independent enough that it should be sent on its own.

quoted

quoted

With this all said -- I noticed iasl prints this statement when trying
to create a node with a lowercase name:
"At least one lower case letter found in NameSeg, ASL is case
insensitive - converting to upper case (GPIO)"

Yes, because it should be in the upper case.

quoted

I wonder if this suggests that adding a call to toupper() to
acpi_fwnode_get_named_child_node would be
an appropriate solution for the node name casing issue....

I dunno. You need to ask in the linux-acpi@ mailing list.
To me this is corner case that can't be easily solved
(because two different specifications treat it differently.

You also need to ask DT people about capital letters there.
And my guts tell me that it's probably also carved in the spec
as "must be lower case" or alike.

FWIW while trying to enable this, at some point I named the I2C and the
GPIO entries "I2C0" and "GPI0" (with the number '0', not the letter
'o'), and it was not working as you would expect.

It is commonly accepted in the ACPI world that the names do not carry
meaning AFAICT, and so I think I agree with Andy's initial comment
regarding using indexes, not names to also fetch the I2C and GPIO nodes.
You can probably have a fallback mechanism for when "i2c" is not
present, or simply check if you are in DT or not and use the names only
if we are in DT.

More and more, after actually seeing and working with ACPI, I suspect that
you both are right. Maybe (hopefully) though, there is some unified way that can
be made to do this, so that individual drivers won't have to directly code for /
be aware of the differences in the firmware languages (at least, that seemed
to be the intent of the fw_node/device api in the first place). Maybe a
`device_get_child_by_name_or_index` (terribly long name) sort of function
might fill in that gap?

I plan to send an email asking this question more generically to ACPI & DT folks
as Andy suggested, so hopefully there will be some ideas.

Thanks a lot to both of you, this will be tremendously helpful to me.

Cheers,
Benjamin

Thanks for testing this out! Glad that ACPI support ended up being worked into
this after all :)