Learn about PCI devices of a host

This commit adds the ability to HostNexus to learn about PCI devices.
For now, we only consider NVIDIA devices for which IOMMU is enabled.

prog/learn_pci.rb

@@ -0,0 +1,50 @@
+# frozen_string_literal: true
+
+class Prog::LearnPci < Prog::Base
+ subject_is :sshable, :vm_host
+
+ REQUIRED_KEYS = ["Slot", "Class", "Vendor", "Device"]
+ PciDeviceRecord = Struct.new(:slot, :device_class, :vendor, :device, :numa_node, :iommu_group) do
+ def self.parse_all(lspci_str)
+ out = []
+ lspci_str.strip.split(/^\n+/).each do |dev_str|
+ dev_h = dev_str.split("\n").map { |e| e.split(":\t") }.to_h
+ fail "BUG: lspci parse failed" unless REQUIRED_KEYS.all? { |s| dev_h.key? s }
+ next unless dev_h.key? "IOMMUGroup"
+ out << PciDeviceRecord.new(dev_h["Slot"], dev_h["Class"], dev_h["Vendor"], dev_h["Device"], dev_h["NUMANode"], dev_h["IOMMUGroup"])
+ end
+ out.freeze
+ end
+ end
+
+ def make_model_instances
+ PciDeviceRecord.parse_all(sshable.cmd("/usr/bin/lspci -vnmm -d 10de::")).map do |rec|
+ PciDevice.new_with_id(
+ vm_host_id: vm_host.id,
+ slot: rec.slot,
+ device_class: rec.device_class,
+ vendor: rec.vendor,
+ device: rec.device,
+ numa_node: rec.numa_node,
+ iommu_group: rec.iommu_group
+ )
+ end
+ end
+
+ label def start
+ make_model_instances.each do |pci|
+ pci.skip_auto_validations(:unique) do
+ pci.insert_conflict(target: [:vm_host_id, :slot],
+ update: {
+ device_class: Sequel[:excluded][:device_class],
+ vendor: Sequel[:excluded][:vendor],
+ device: Sequel[:excluded][:device],
+ numa_node: Sequel[:excluded][:numa_node],
+ iommu_group: Sequel[:excluded][:iommu_group]
+ }).save_changes
+ end
+ end
+
+ pop("created PciDevice records")
+ end
+end

prog/vm/host_nexus.rb

@@ -56,6 +56,7 @@ def before_run
  bud Prog::LearnArch
  bud Prog::LearnCores
  bud Prog::LearnStorage
+ bud Prog::LearnPci
  bud Prog::InstallDnsmasq
  bud Prog::SetupSysstat
  bud Prog::SetupNftables

spec/prog/learn_pci_spec.rb

@@ -0,0 +1,111 @@
+# frozen_string_literal: true
+
+require_relative "../model/spec_helper"
+
+RSpec.describe Prog::LearnPci do
+ describe "#make_model_instances" do
+ let(:nvidia_gpu_with_audio) do
+ end
+
+ it "exits, saving model instances" do
+ vmh = Prog::Vm::HostNexus.assemble("::1").subject
+ lp = described_class.new(Strand.new(stack: [{"subject_id" => vmh.id}]))
+ expect(lp.sshable).to receive(:cmd).with("/usr/bin/lspci -vnmm -d 10de::").and_return(<<EOS)
+Slot: 01:00.0
+Class: 0300
+Vendor: 10de
+Device: 27b0
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+NUMANode: 1
+IOMMUGroup: 13
+
+Slot: 01:00.1
+Class: 0403
+Vendor: 10de
+Device: 22bc
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+IOMMUGroup: 13
+EOS
+ expect { lp.start }.to exit({"msg" => "created PciDevice records"}).and change {
+ PciDevice.map { {vm_host_id: _1.vm_host_id, slot: _1.slot, device_class: _1.device_class, vendor: _1.vendor, device: _1.device, numa_node: _1.numa_node, iommu_group: _1.iommu_group, vm_id: _1.vm_id} }.sort_by { _1[:slot] }
+ }.from(
+ []
+ ).to(
+ [
+ {vm_host_id: vmh.id, slot: "01:00.0", device_class: "0300", vendor: "10de", device: "27b0", numa_node: 1, iommu_group: 13, vm_id: nil},
+ {vm_host_id: vmh.id, slot: "01:00.1", device_class: "0403", vendor: "10de", device: "22bc", numa_node: nil, iommu_group: 13, vm_id: nil}
+ ]
+ )
+ end
+
+ it "exits, updating existing model instances" do
+ vmh = Prog::Vm::HostNexus.assemble("::1").subject
+ lp = described_class.new(Strand.new(stack: [{"subject_id" => vmh.id}]))
+ PciDevice.create_with_id(vm_host_id: vmh.id, slot: "01:00.0", device_class: "dc", vendor: "vd", device: "dv", numa_node: 0, iommu_group: 3)
+ expect(lp.sshable).to receive(:cmd).with("/usr/bin/lspci -vnmm -d 10de::").and_return(<<EOS)
+Slot: 01:00.0
+Class: 0300
+Vendor: 10de
+Device: 27b0
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+IOMMUGroup: 13
+
+Slot: 01:00.1
+Class: 0403
+Vendor: 10de
+Device: 22bc
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+IOMMUGroup: 13
+EOS
+ expect { lp.start }.to exit({"msg" => "created PciDevice records"}).and change {
+ PciDevice.map { {vm_host_id: _1.vm_host_id, slot: _1.slot, device_class: _1.device_class, vendor: _1.vendor, device: _1.device, numa_node: _1.numa_node, iommu_group: _1.iommu_group, vm_id: _1.vm_id} }.sort_by { _1[:slot] }
+ }.from(
+ [{vm_host_id: vmh.id, slot: "01:00.0", device_class: "dc", vendor: "vd", device: "dv", numa_node: 0, iommu_group: 3, vm_id: nil}]
+ ).to(
+ [
+ {vm_host_id: vmh.id, slot: "01:00.0", device_class: "0300", vendor: "10de", device: "27b0", numa_node: nil, iommu_group: 13, vm_id: nil},
+ {vm_host_id: vmh.id, slot: "01:00.1", device_class: "0403", vendor: "10de", device: "22bc", numa_node: nil, iommu_group: 13, vm_id: nil}
+ ]
+ )
+ end
+
+ it "ignores devices without iommu group" do
+ vmh = Prog::Vm::HostNexus.assemble("::1").subject
+ lp = described_class.new(Strand.new(stack: [{"subject_id" => vmh.id}]))
+ expect(lp.sshable).to receive(:cmd).with("/usr/bin/lspci -vnmm -d 10de::").and_return(<<EOS)
+Slot: 01:00.0
+Class: 0300
+Vendor: 10de
+Device: 27b0
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+EOS
+ expect(lp.make_model_instances).to eq([])
+ end
+
+ it "can raise a data parse error" do
+ vmh = Prog::Vm::HostNexus.assemble("::1").subject
+ lp = described_class.new(Strand.new(stack: [{"subject_id" => vmh.id}]))
+ expect(lp.sshable).to receive(:cmd).with("/usr/bin/lspci -vnmm -d 10de::").and_return(<<EOS)
+Slot: 01:00.0
+Class: 0300
+Device: 27b0
+SVendor: 10de
+SDevice: 16fa
+Rev: a1
+IOMMUGroup: 13
+EOS
+
+ expect { lp.make_model_instances }.to raise_error RuntimeError, "BUG: lspci parse failed"
+ end
+ end
+end

spec/prog/vm/host_nexus_spec.rb

@@ -106,6 +106,7 @@
  Prog::LearnArch,
  Prog::LearnCores,
  Prog::LearnStorage,
+ Prog::LearnPci,
  Prog::InstallDnsmasq,
  Prog::SetupSysstat,
  Prog::SetupNftables