uefi: Add device path generation for discovered devices in a PciTree

[seijikun]

Initial implementation for device path generation of discovered pci devices in a PciTree instance.
Unfortunately, it's a bit clunky to use, since you have to query the DevicePath corresponding to your PciRootBridgeIo instance yourself and pass that in:

let mut pci_proto = get_open_protocol::<PciRootBridgeIo>(pci_handle);
let root_device_path = get_open_protocol::<DevicePath>(pci_handle);

let pci_tree = pci_proto.enumerate().unwrap();
for addr in pci_tree {
    // v Would be nicer, if I didn't have to pass root DevicePath in manually here v
    let device_path = pci_tree.device_path(&root_device_path, addr).unwrap();
}

I tried getting the DevicePath of the PciRootBridgeIo instance within PciRootBridgeIo::enumerate() and pass that on to the PciTree, but I couldn't find a way to get the DevicePath instance - other than ugly workarounds.

Best consumed Commit-By-Commit. Contains #1830..

Steps to Undraft

• merge and rebase onto uefi: Refactor PciRootBridgeIo::enumerate() with tree-topology information #1830

Checklist

• Sensible git history (for example, squash "typo" or "fix" commits). See the Rewriting History guide for help.
• Update the changelog (if necessary)

[phip1611]

LGTM with two nits!

PS: You do not have to fix it here but do you have an opinion on #1817? I'd like to hear a direction that we should go for (Box<DevicePath> vs PoolDevicePath)

[seijikun]

You do not have to fix it here but do you have an opinion on #1817? I'd like to hear a direction that we should go for (Box vs PoolDevicePath)

To be completely honest with you, I don't really understand the difference. I was under the impression, that PoolDevicePath is just a typedef of Box<DevicePath>. Is that not correct?

[phip1611]

You do not have to fix it here but do you have an opinion on #1817? I'd like to hear a direction that we should go for (Box vs PoolDevicePath)

To be completely honest with you, I don't really understand the difference. I was under the impression, that PoolDevicePath is just a typedef of Box<DevicePath>. Is that not correct?

Nope, it is a wrapper for PoolAllocation. https://docs.rs/uefi/latest/src/uefi/proto/device_path/mod.rs.html#137

[seijikun]

Nope, it is a wrapper for PoolAllocation. https://docs.rs/uefi/latest/src/uefi/proto/device_path/mod.rs.html#137

Yes. But isn't alloc in uefi also just using pool allocations under the hood, conceptually?

[phip1611]

LGTM, thanks for your contribution!

We are currently inconsistent with Box<[u8]> and PoolDevicePath - no need to clean that up.

One final question: When I execute the integration test, it prints

...
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@069:  \- Bus: 01
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [00, 1f, 00]: vendor=8086, device=2918, class=06, subclass=01 - PciRoot(0x0)/Pci(0x1F,0x0)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [00, 1f, 02]: vendor=8086, device=2922, class=01, subclass=06 - PciRoot(0x0)/Pci(0x1F,0x2)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [00, 1f, 03]: vendor=8086, device=2930, class=0C, subclass=05 - PciRoot(0x0)/Pci(0x1F,0x3)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [01, 00, 00]: vendor=104C, device=8232, class=06, subclass=04 - PciRoot(0x0)/Pci(0x7,0x0)/Pci(0x0,0x0)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@069:  \- Bus: 02
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [02, 00, 00]: vendor=104C, device=8233, class=06, subclass=04 - PciRoot(0x0)/Pci(0x7,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@069:  \- Bus: 03
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [02, 01, 00]: vendor=104C, device=8233, class=06, subclass=04 - PciRoot(0x0)/Pci(0x7,0x0)/Pci(0x0,0x0)/Pci(0x1,0x0)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@069:  \- Bus: 04
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [03, 00, 00]: vendor=1AF4, device=1048, class=01, subclass=00 - PciRoot(0x0)/Pci(0x7,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)/Pci(0x0,0x0)
[ INFO]: uefi-test-runner/src/proto/pci/root_bridge.rs@064: PCI Device: [04, 00, 00]: vendor=1AF4, device=1048, class=01, subclass=00 - PciRoot(0x0)/Pci(0x7,0x0)/Pci(0x0,0x0)/Pci(0x1,0x0)/Pci(0x0,0x0)
[ INFO]: uefi-test-runner/src/proto/pi/mod.rs@004: Testing Platform Initialization protocols

I don't understand the output. When you are about to enumerate bus 3, the device says it is at bus 2? And if you enumerate Bus 4, the first device says it is as bus 3? Please explain - or is therea bug?

Not sure about the random Windows CI failure, tho.

[seijikun]

Ah yes, the logging may be a bit confusing.

        let pci_tree = pci_proto.enumerate().unwrap();
        for addr in pci_tree.iter().cloned() {
                // ...
        }

basically performs a breadth-first-search. Meaning that you will first get all devices in bus 0 before going to bus 1.
At the bottom of the loop, I print the subsequent bus numbers (if any) for every iterated device with this:

            for child_bus in pci_tree.child_bus_of_iter(addr) {
                log::info!(" \\- Bus: {child_bus:02x}");
            }

So the log line \- Bus: 03 is basically only saying: "The bus 03 that will come sometime later is connected to me"