Re: [PATCH v3 3/3] selftests: net: add a virtio_net deadlock selftest

[Bui Quang Minh <minhquangbui99@gmail.com>]

On 4/16/25 14:46, Jason Wang wrote:
> On Wed, Apr 16, 2025 at 2:54 PM Bui Quang Minh <minhquangbui99@gmail.com> wrote:
>> On 4/16/25 11:27, Jakub Kicinski wrote:
>>> On Tue, 15 Apr 2025 14:43:41 +0700 Bui Quang Minh wrote:
>>>> +def setup_xsk(cfg, xdp_queue_id = 0) -> bkg:
>>>> +    # Probe for support
>>>> +    xdp = cmd(f'{cfg.net_lib_dir / "xdp_helper"} - -', fail=False)
>>>> +    if xdp.ret == 255:
>>>> +        raise KsftSkipEx('AF_XDP unsupported')
>>>> +    elif xdp.ret > 0:
>>>> +        raise KsftFailEx('unable to create AF_XDP socket')
>>>> +
>>>> +    return bkg(f'{cfg.net_lib_dir / "xdp_helper"} {cfg.ifindex} {xdp_queue_id}',
>>>> +               ksft_wait=3)
>>>> +
>>>> +def check_xdp_bind(cfg):
>>>> +    ip(f"link set dev %s xdp obj %s sec xdp" %
>>>> +       (cfg.ifname, cfg.net_lib_dir / "xdp_dummy.bpf.o"))
>>>> +    ip(f"link set dev %s xdp off" % cfg.ifname)
>>>> +
>>>> +def check_rx_resize(cfg, queue_size = 128):
>>>> +    rx_ring = _get_rx_ring_entries(cfg)
>>>> +    ethtool(f"-G %s rx %d" % (cfg.ifname, queue_size))
>>>> +    ethtool(f"-G %s rx %d" % (cfg.ifname, rx_ring))
>>> Unfortunately this doesn't work on a basic QEMU setup:
>>>
>>> # ethtool -G eth0 rx 128
>>> [   15.680655][  T287] virtio_net virtio2 eth0: resize rx fail: rx queue index: 0 err: -2
>>> netlink error: No such file or directory
>>>
>>> Is there a way to enable more capable virtio_net with QEMU?
> What's the qemu command line and version?
>
> Resize depends on queue_reset which should be supported from Qemu 7.2
>
>> I guess that virtio-pci-legacy is used in your setup.
> Note that modern devices are used by default.
>
>> Here is how I setup virtio-net with Qemu
>>
>>       -netdev tap,id=hostnet1,vhost=on,script=$NETWORK_SCRIPT,downscript=no \
>>       -device
>> virtio-net-pci,netdev=hostnet1,iommu_platform=on,disable-legacy=on \
>>
>> The iommu_platform=on is necessary to make vring use dma API which is a
>> requirement to enable xsk_pool in virtio-net (XDP socket will be in
>> zerocopy mode for this case). Otherwise, the XDP socket will fallback to
>> copy mode, xsk_pool is not enabled in virtio-net that makes the
>> probability to reproduce bug to be very small. Currently, when you don't
>> have iommu_platform=on, you can pass the test even before the fix, so I
>> think I will try to harden the selftest to make it return skip in this case.
> I would like to keep the resize test as it doesn't require iommu_platform.

Okay, in next version I will force the XDP socket binding to zerocopy to 
setup xsk_pool. When the binding fails, 2 tests still run but I will 
print a warning message.

Thanks,
Quang Minh.