ci: Add test_vfio to the baremetal CI

Since it's not possible to run the integration test test_vfio on Azure
at the moment (because of some nested virtualization issues), we can
temporarily run it on the baremetal CI where we already run some VFIO
tests.

Signed-off-by: Sebastien Boeuf <sebastien.boeuf@intel.com>
This commit is contained in:
Sebastien Boeuf 2022-08-03 17:28:53 +02:00 committed by Rob Bradford
parent 9b4940f154
commit 3b0f6247b7
2 changed files with 305 additions and 2 deletions

View File

@ -8,6 +8,33 @@ process_common_args "$@"
# For now these values are default for kvm
features=""
WORKLOADS_DIR="$HOME/workloads"
FOCAL_OS_IMAGE_NAME="focal-server-cloudimg-amd64-custom-20210609-0.raw"
FOCAL_OS_IMAGE="$WORKLOADS_DIR/$FOCAL_OS_IMAGE_NAME"
FW="$WORKLOADS_DIR/hypervisor-fw"
VMLINUX_IMAGE="$WORKLOADS_DIR/vmlinux"
BLK_IMAGE="$WORKLOADS_DIR/blk.img"
MNT_DIR="mount_image"
rm -rf $BLK_IMAGE
pushd $WORKLOADS_DIR
fallocate -l 16M $BLK_IMAGE
mkfs.ext4 -j $BLK_IMAGE
mkdir $MNT_DIR
sudo mount -t ext4 $BLK_IMAGE $MNT_DIR
sudo bash -c "echo bar > $MNT_DIR/foo" || exit 1
sudo umount $BLK_IMAGE
rm -r $MNT_DIR
popd
VFIO_DIR="$WORKLOADS_DIR/vfio"
VFIO_DISK_IMAGE="$WORKLOADS_DIR/vfio.img"
rm -rf $VFIO_DIR $VFIO_DISK_IMAGE
mkdir -p $VFIO_DIR
cp $FOCAL_OS_IMAGE $VFIO_DIR
cp $FW $VFIO_DIR
cp $VMLINUX_IMAGE $VFIO_DIR || exit 1
BUILD_TARGET="$(uname -m)-unknown-linux-${CH_LIBC}"
CFLAGS=""
TARGET_CC=""
@ -18,10 +45,23 @@ fi
cargo build --all --release $features --target $BUILD_TARGET
strip target/$BUILD_TARGET/release/cloud-hypervisor
strip target/$BUILD_TARGET/release/ch-remote
# We always copy a fresh version of our binary for our L2 guest.
cp target/$BUILD_TARGET/release/cloud-hypervisor $VFIO_DIR
cp target/$BUILD_TARGET/release/ch-remote $VFIO_DIR
# test_vfio rely on hugepages
echo 6144 | sudo tee /proc/sys/vm/nr_hugepages
sudo chmod a+rwX /dev/hugepages
export RUST_BACKTRACE=1
time cargo test $features "vfio::$test_filter" -- --test-threads=1 ${test_binary_args[*]}
time cargo test $features "vfio::test_vfio" -- ${test_binary_args[*]}
RES=$?
if [ $RES -eq 0 ]; then
time cargo test $features "vfio::test_nvidia" -- --test-threads=1 ${test_binary_args[*]}
RES=$?
fi
exit $RES

View File

@ -7681,6 +7681,269 @@ mod sgx {
mod vfio {
use crate::*;
#[test]
// The VFIO integration test starts cloud-hypervisor guest with 3 TAP
// backed networking interfaces, bound through a simple bridge on the host.
// So if the nested cloud-hypervisor succeeds in getting a directly
// assigned interface from its cloud-hypervisor host, we should be able to
// ssh into it, and verify that it's running with the right kernel command
// line (We tag the command line from cloud-hypervisor for that purpose).
// The third device is added to validate that hotplug works correctly since
// it is being added to the L2 VM through hotplugging mechanism.
// Also, we pass-through a vitio-blk device to the L2 VM to test the 32-bit
// vfio device support
fn test_vfio() {
setup_vfio_network_interfaces();
let focal = UbuntuDiskConfig::new(FOCAL_IMAGE_NAME.to_string());
let guest = Guest::new_from_ip_range(Box::new(focal), "172.18", 0);
let mut workload_path = dirs::home_dir().unwrap();
workload_path.push("workloads");
let kernel_path = direct_kernel_boot_path();
let mut vfio_path = workload_path.clone();
vfio_path.push("vfio");
let mut cloud_init_vfio_base_path = vfio_path.clone();
cloud_init_vfio_base_path.push("cloudinit.img");
// We copy our cloudinit into the vfio mount point, for the nested
// cloud-hypervisor guest to use.
rate_limited_copy(
&guest.disk_config.disk(DiskType::CloudInit).unwrap(),
&cloud_init_vfio_base_path,
)
.expect("copying of cloud-init disk failed");
let mut vfio_disk_path = workload_path.clone();
vfio_disk_path.push("vfio.img");
// Create the vfio disk image
let output = Command::new("mkfs.ext4")
.arg("-d")
.arg(vfio_path.to_str().unwrap())
.arg(vfio_disk_path.to_str().unwrap())
.arg("2g")
.output()
.unwrap();
if !output.status.success() {
eprintln!("{}", String::from_utf8_lossy(&output.stderr));
panic!("mkfs.ext4 command generated an error");
}
let mut blk_file_path = workload_path;
blk_file_path.push("blk.img");
let vfio_tap0 = "vfio-tap0";
let vfio_tap1 = "vfio-tap1";
let vfio_tap2 = "vfio-tap2";
let vfio_tap3 = "vfio-tap3";
let mut child = GuestCommand::new(&guest)
.args(&["--cpus", "boot=4"])
.args(&["--memory", "size=2G,hugepages=on,shared=on"])
.args(&["--kernel", kernel_path.to_str().unwrap()])
.args(&[
"--disk",
format!(
"path={}",
guest.disk_config.disk(DiskType::OperatingSystem).unwrap()
)
.as_str(),
format!(
"path={}",
guest.disk_config.disk(DiskType::CloudInit).unwrap()
)
.as_str(),
format!("path={}", vfio_disk_path.to_str().unwrap()).as_str(),
format!("path={},iommu=on", blk_file_path.to_str().unwrap()).as_str(),
])
.args(&[
"--cmdline",
format!(
"{} kvm-intel.nested=1 vfio_iommu_type1.allow_unsafe_interrupts",
DIRECT_KERNEL_BOOT_CMDLINE
)
.as_str(),
])
.args(&[
"--net",
format!("tap={},mac={}", vfio_tap0, guest.network.guest_mac).as_str(),
format!(
"tap={},mac={},iommu=on",
vfio_tap1, guest.network.l2_guest_mac1
)
.as_str(),
format!(
"tap={},mac={},iommu=on",
vfio_tap2, guest.network.l2_guest_mac2
)
.as_str(),
format!(
"tap={},mac={},iommu=on",
vfio_tap3, guest.network.l2_guest_mac3
)
.as_str(),
])
.capture_output()
.spawn()
.unwrap();
thread::sleep(std::time::Duration::new(30, 0));
let r = std::panic::catch_unwind(|| {
guest.ssh_command_l1("sudo systemctl start vfio").unwrap();
thread::sleep(std::time::Duration::new(120, 0));
// We booted our cloud hypervisor L2 guest with a "VFIOTAG" tag
// added to its kernel command line.
// Let's ssh into it and verify that it's there. If it is it means
// we're in the right guest (The L2 one) because the QEMU L1 guest
// does not have this command line tag.
assert_eq!(
guest
.ssh_command_l2_1("grep -c VFIOTAG /proc/cmdline")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
1
);
// Let's also verify from the second virtio-net device passed to
// the L2 VM.
assert_eq!(
guest
.ssh_command_l2_2("grep -c VFIOTAG /proc/cmdline")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
1
);
// Check the amount of PCI devices appearing in L2 VM.
assert_eq!(
guest
.ssh_command_l2_1("ls /sys/bus/pci/devices | wc -l")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
8,
);
// Check both if /dev/vdc exists and if the block size is 16M in L2 VM
assert_eq!(
guest
.ssh_command_l2_1("lsblk | grep vdc | grep -c 16M")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
1
);
// Hotplug an extra virtio-net device through L2 VM.
guest
.ssh_command_l1(
"echo 0000:00:09.0 | sudo tee /sys/bus/pci/devices/0000:00:09.0/driver/unbind",
)
.unwrap();
guest
.ssh_command_l1("echo 0000:00:09.0 | sudo tee /sys/bus/pci/drivers/vfio-pci/bind")
.unwrap();
let vfio_hotplug_output = guest
.ssh_command_l1(
"sudo /mnt/ch-remote \
--api-socket=/tmp/ch_api.sock \
add-device path=/sys/bus/pci/devices/0000:00:09.0,id=vfio123",
)
.unwrap();
assert!(vfio_hotplug_output.contains("{\"id\":\"vfio123\",\"bdf\":\"0000:00:08.0\"}"));
thread::sleep(std::time::Duration::new(10, 0));
// Let's also verify from the third virtio-net device passed to
// the L2 VM. This third device has been hotplugged through the L2
// VM, so this is our way to validate hotplug works for VFIO PCI.
assert_eq!(
guest
.ssh_command_l2_3("grep -c VFIOTAG /proc/cmdline")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
1
);
// Check the amount of PCI devices appearing in L2 VM.
// There should be one more device than before, raising the count
// up to 9 PCI devices.
assert_eq!(
guest
.ssh_command_l2_1("ls /sys/bus/pci/devices | wc -l")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
9,
);
// Let's now verify that we can correctly remove the virtio-net
// device through the "remove-device" command responsible for
// unplugging VFIO devices.
guest
.ssh_command_l1(
"sudo /mnt/ch-remote \
--api-socket=/tmp/ch_api.sock \
remove-device vfio123",
)
.unwrap();
thread::sleep(std::time::Duration::new(10, 0));
// Check the amount of PCI devices appearing in L2 VM is back down
// to 8 devices.
assert_eq!(
guest
.ssh_command_l2_1("ls /sys/bus/pci/devices | wc -l")
.unwrap()
.trim()
.parse::<u32>()
.unwrap_or_default(),
8,
);
// Perform memory hotplug in L2 and validate the memory is showing
// up as expected. In order to check, we will use the virtio-net
// device already passed through L2 as a VFIO device, this will
// verify that VFIO devices are functional with memory hotplug.
assert!(guest.get_total_memory_l2().unwrap_or_default() > 480_000);
guest
.ssh_command_l2_1(
"sudo bash -c 'echo online > /sys/devices/system/memory/auto_online_blocks'",
)
.unwrap();
guest
.ssh_command_l1(
"sudo /mnt/ch-remote \
--api-socket=/tmp/ch_api.sock \
resize --memory=1073741824",
)
.unwrap();
assert!(guest.get_total_memory_l2().unwrap_or_default() > 960_000);
});
let _ = child.kill();
let output = child.wait_with_output().unwrap();
cleanup_vfio_network_interfaces();
handle_child_output(r, &output);
}
fn test_nvidia_card_memory_hotplug(hotplug_method: &str) {
let hirsute = UbuntuDiskConfig::new(HIRSUTE_NVIDIA_IMAGE_NAME.to_string());
let guest = Guest::new(Box::new(hirsute));