Skip to content

Latest commit

 

History

History
76 lines (44 loc) · 4.4 KB

PCI-Construction-Method.md

File metadata and controls

76 lines (44 loc) · 4.4 KB

What is this?

This, kind user, is a markdown file where we explain how to properly construct the PCI path of a PCI device on all platforms.

Enjoy!

Linux

The only situation where it's applicable (as far as I know) to construct a PCI path via PCIDebug is on Linux. There, we would want to look at the amount of “path components” the device's ACPI path contains past \_SB.PCI0; so let's say we had \_SB.PCI0.GFX0, this would mean that there should only be a singular PCI path component (excluding PciRoot()) in the final path. Say that this device's PCIDebug was 0000:00:02.0, we would want to set PciRoot(...) (where the value inside of is the hex representation of the domain – 0000 -> 0x0), then we'd take a look at the device ID (02) and function value (.0).

From here, we construct the following PCI path: PciRoot(0x0)/Pci(0x2,0x0)

PciRoot(0x0)/Pci(0x2,0x0)
        ||       ||  | _ _ _ _ Function (0 => 0x0)
        ||       ||
        ||       | _ _ _ _ Device ID (02 => 0x2)
        | _ _ _ _ Domain value (0000 => 0x0)

However, if the ACPI path has more than a singular “path component,” we'd want to traverse /sys/bus/pci/devices and look for an entry that matches the device's PCIDebug value.

So let's take for example: \_SB.PCI0.PEG0.PEGP

Its PCIDebug value is 0000:00:01.0, initially.

As previously mentioned, we'll traverse /sys/bus/pci/devices and look for a match. In this example, we have found it: /sys/bus/pci/devices/0000:00:01.0; inside of this directory, we may or may not find a few other directories which also have their own PCIDebug values, they'll also, most likely, include pcie in their directory name.

Looking back at the amount of path components our ACPI path (\_SB.PCI0.PEG0.PEGP) has, we can see that it's 2 (after removing \_SB.PCI0), and since we already have the first PCIDebug value, we'd want to take the first entry inside of the parent entry's directory (/sys/bus/pci/devices/0000:00:01.0); from where we may find the following:

  • 0000:00:01.0:pcie002
  • 0000:00:01.0:pcie010

We'll want to take the first entry found (*:pcie002), since we need only 1 additional PCIDebug value. We extract the PCIDebug by slicing off :pcie002, and that'll be the second PCI path component. Which finally yields:

PciRoot(0x0)/Pci(0x1,0x0)/Pci(0x0,0x0)

Windows

On Windows it's pretty much just a matter of properly slicing a string and splitting it – PCIROOT(0)#PCI(0100)#PCI(0000) -> PciRoot(0x0)/Pci(0x1,0x0)/Pci(0x0,0x0)

Logic behind the Windows implementation

macOS

On macOS, it's slightly more convoluted, but ultimately equally as accurate and proper.

Logic behind the macOS implementation (special thank we to DhinakG and rest of the Dortania team for allowing us to copy over this implementation into OCSysInfo)

Before this more improved and proper logic on macOS, we previously had looked at its “location in plane” value.

Here's an example of that: IOACPIPlane:/_SB/PCI0@0/HDAS@1f0003

We'd just remove IOACPIPlane:/, and then split by /, which would yield:

  • ["_SB", "PCI0@0", "HDAS@1f0003"]

We'd then traverse this list and first take care of _SB, as you can see, it doesn't include a @ in its value, which is normal, but the rest of these components include their domain (PCI0 in this example) or device id + function (HDAS in this example), so we'd just append \_SB to the “formatted list,” and then take care of the rest if it has the @ value.

If it does include the @ value, we'd split on it, append the first value (e.g. PCI0) to the final ACPI path, and, before using bitshifts to extract device id and function values, check if the ACPI value is PCI0, in which case it'd be our PciRoot(), otherwise, it'd be Pci()

After we encounter HDAS@1f0003, from here, we of course split on @ and append the former value to the ACPI path, and deal with the latter value with bitshifts:

acpi, pcid = val.split("@")

pcid = int(pcid, 16)

device_id = hex(pcid >> 16 & 0xFFFF) #  -> 0x1f
function  = hex(pcid & 0xFFFF)       #  -> 0x3

Here's what's happening with the bitshifts, in case you wanted to know: