More predictable network interface names for PCI(e) devices

[jclehner]

Submission type

• Bug report
• Request for enhancement (RFE)

Used distribution

Ubuntu 15.10

Bugreport

I recently removed the GPU from my system, and noticed that my network interface names changed (enp4s1 -> enp3s1 and enp6s0 -> enp5s0 in my case). I know I'm late to the game, since this issue is similar to #563 but I'm also proposing a portable solution.

This is my pci configuration with a dedicated GPU (network interfaces enp4s1 and enp6s0):

$ lspci -vtnn
-[0000:00]-+-00.0  Intel Corporation Sky Lake Host Bridge/DRAM Registers [8086:191f]
           +-01.0-[01]----00.0  NVIDIA Corporation G84 [GeForce 8600 GTS] [10de:0400]
           +-02.0  Intel Corporation Sky Lake Integrated Graphics [8086:1912]
           +-14.0  Intel Corporation Sunrise Point-H USB 3.0 xHCI Controller [8086:a12f]
           +-14.2  Intel Corporation Sunrise Point-H Thermal subsystem [8086:a131]
           +-16.0  Intel Corporation Sunrise Point-H CSME HECI #1 [8086:a13a]
           +-17.0  Intel Corporation Sunrise Point-H SATA controller [AHCI mode] [8086:a102]
           +-1b.0-[02]--
           +-1b.2-[03-04]----00.0-[04]--+-00.0  Promise Technology, Inc. PDC20267 (FastTrak100/Ultra100) [105a:4d30]
           |                            \-01.0  Realtek Semiconductor Co., Ltd. RTL8169 PCI Gigabit Ethernet Controller [10ec:8169]
           +-1c.0-[05]----00.0  ASMedia Technology Inc. ASM1142 USB 3.1 Host Controller [1b21:1242]
           +-1c.3-[06]----00.0  Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller [10ec:8168]
           +-1c.4-[07]--
           +-1d.0-[08]--
           +-1d.4-[09]--
           +-1f.0  Intel Corporation Sunrise Point-H LPC Controller [8086:a145]
           +-1f.2  Intel Corporation Sunrise Point-H PMC [8086:a121]
           +-1f.3  Intel Corporation Sunrise Point-H HD Audio [8086:a170]
           \-1f.4  Intel Corporation Sunrise Point-H SMBus [8086:a123]

And this is the same configuration with the dedicated GPU removed (enp3s1 and enp5s0):

$ lspci -vtnn
-[0000:00]-+-00.0  Intel Corporation Sky Lake Host Bridge/DRAM Registers [8086:191f]
           +-02.0  Intel Corporation Sky Lake Integrated Graphics [8086:1912]
           +-14.0  Intel Corporation Sunrise Point-H USB 3.0 xHCI Controller [8086:a12f]
           +-14.2  Intel Corporation Sunrise Point-H Thermal subsystem [8086:a131]
           +-16.0  Intel Corporation Sunrise Point-H CSME HECI #1 [8086:a13a]
           +-17.0  Intel Corporation Sunrise Point-H SATA controller [AHCI mode] [8086:a102]
           +-1b.0-[01]--
           +-1b.2-[02-03]----00.0-[03]--+-00.0  Promise Technology, Inc. PDC20267 (FastTrak100/Ultra100) [105a:4d30]
           |                            \-01.0  Realtek Semiconductor Co., Ltd. RTL8169 PCI Gigabit Ethernet Controller [10ec:8169]
           +-1c.0-[04]----00.0  ASMedia Technology Inc. ASM1142 USB 3.1 Host Controller [1b21:1242]
           +-1c.3-[05]----00.0  Realtek Semiconductor Co., Ltd. RTL8111/8168/8411 PCI Express Gigabit Ethernet Controller [10ec:8168]
           +-1c.4-[06]--
           +-1d.0-[07]--
           +-1d.4-[08]--
           +-1f.0  Intel Corporation Sunrise Point-H LPC Controller [8086:a145]
           +-1f.2  Intel Corporation Sunrise Point-H PMC [8086:a121]
           +-1f.3  Intel Corporation Sunrise Point-H HD Audio [8086:a170]
           \-1f.4  Intel Corporation Sunrise Point-H SMBus [8086:a123]

As is obvious from the lspci output, the bus numbers of all devices are shifted by 1 if the dedicated GPU is installed, thus explaining the changed interface names. One thing however remains constant: the BDF of the PCI root ports: 00:1c.4 for enp{5,6}s0, and 00:1b.2 for enp{3,4}s1, so my proposal is this:

If the interface names used some form of notation relative to their respective PCI root ports, wouldn't this be a more elegant solution (aside from the resulting interface name that is, but heh, it's more predictable)? For example, enr28f4p0s0 could be used to denote a network interface behind the PCI root port with slot 28 (==0x1c) and function 4. The p0 part in this case is to be interpreted relative to the bridge's secondary bus, i.e. if the bridge is configured with a secondary bus of 5, r28f4p0 is the same as p5.

Please comment.

Regards
Joseph

[poettering]

@teg?

[kalvdans]

I have a similar case as @jclehner. With CAN adapter in a PCIe slot bumps the bus numbers of the two network devices one step.

           +-1b.0-[02]----00.0  IXXAT Automation GmbH Device [1bee:0002]
           +-1c.0-[03]----00.0  Aquantia Corp. Device [1d6a:00b1]
           +-1d.0-[04]----00.0  Intel Corporation I210 Gigabit Network Connection [8086:1533]

Without can adapter:

           +-1c.0-[02]----00.0  Aquantia Corp. Device [1d6a:00b1]
           +-1d.0-[03]----00.0  Intel Corporation I210 Gigabit Network Connection [8086:1533]

biosdevname gives deterministic names p4p1 and p8p1 but I rather not depend on bios for the naming.

[poettering]

Let's close this. The bug is 7 years old, and this is not really actionable to us anyway: if firmware re-numerates supposedly stable slot ids, there's nothing we can do. please complain to manufacturer.

sorry