sysfs: add SR-IOV VF PCI address and NUMA node helpers for net interfaces#795
sysfs: add SR-IOV VF PCI address and NUMA node helpers for net interfaces#795aharivel wants to merge 1 commit into
Conversation
|
Why is this not part of |
|
@SuperQ I followed the precedent of You're right that net_class.go is the natural home. Both functions read attributes from the underlying PCI device of a net interface (device/numa_node, device/virtfn), which is exactly the kind of thing net_class.go already handles. I can move them there and drop net_class_sriov.go. That said — would you prefer they stay grouped together (e.g. with a comment marking them as SR-IOV related), or just inline them with the rest of the net class helpers? |
|
Hmm, looking into this a bit more. The |
|
Hmm ok cleaner approach — agreed. A For virtfnX resolution, I'd read those symlinks from the resolved PCI device path. |
Add two new functions to support SR-IOV network VF metrics collection: - NetClassPCIDevice(iface string) (*PciDevice, error) in net_class.go: Resolves /sys/class/net/<iface>/device symlink and returns the backing PciDevice using the existing PCI device parser. This gives callers access to all PCI device properties (NumaNode, link speed, SR-IOV fields, etc.). - PciDeviceVFAddress(device *PciDevice, vfIndex uint32) (string, error) in pci_device.go: Returns the PCI BDF address of a Virtual Function by resolving the virtfn symlink at /sys/bus/pci/devices/<bdf>/virtfn<n>. Test fixtures added under testdata/fixtures/sys/class/net/enp3s0f0/ using the existing 0000:a2:00.0 PCI device fixture. Signed-off-by: Anthony Harivel <aharivel@redhat.com>
|
@SuperQ Ok so I've reworked the patch in a way that takes your feedback into consideration. I've also tested this morning the end goal of this patch (e.g adding SR-IOV network Virtual Function statistics in node_exporter prometheus/node_exporter#3544 ) on my server with E810 nic and the result looks as expected (i.e PCI address are good, numa node are also good): |
|
Hi @SuperQ just a gentle ping on this PR, would you have a moment to take a look when you get a chance? Thanks! |
|
Hi @discordianfish |
|
Hi, is there anything I can do to help review this patch ? |
SuperQ
left a comment
SuperQ
left a comment
There was a problem hiding this comment.
Looks good, small issue with the Go docs.
| @@ -97,6 +97,18 @@ func (fs FS) NetClassByIface(devicePath string) (*NetClassIface, error) { | |||
| } | |||
|
|
|||
| // NetClass returns info for all net interfaces (iface) read from /sys/class/net/<iface>. | |||
There was a problem hiding this comment.
It looks like you've misplaced this doc comment. It needs to be attached to func (fs FS) NetClass() below.
2 participants
Add two helpers to the sysfs package for SR-IOV Physical Function (PF) network interfaces:
NetClassVFPCIAddress(iface string, vfIndex uint32) (string, error)Resolves the PCI BDF address of a Virtual Function by reading the sysfs
virtfn symlink at
/sys/class/net/<iface>/device/virtfn<vfIndex>.Returns the BDF address e.g.
0000:65:01.0.NetClassNumaNode(iface string) (int, error)Returns the NUMA node of the PCI device backing the interface, read from
/sys/class/net/<iface>/device/numa_node.Returns -1 if the device is not NUMA-aware (kernel convention).
These are intended to support SR-IOV VF collectors (e.g. in node_exporter) that need to correlate VF traffic metrics with workloads referencing VFs by PCI address or NUMA topology (e.g. OpenStack Nova, libvirt, DPDK).
Test fixtures added under
testdata/fixtures/sys/class/net/enp3s0f0/.