[RFC] test/eal_flags: suppress vdev_netvsc auto-probe

[RFC] test/eal_flags: suppress vdev_netvsc auto-probe

[Stephen Hemminger]

On Hyper-V systems (including Azure-hosted CI), the vdev_netvsc
driver auto-injects itself during vdev bus scan. This interferes
with the vdev flag test causing the nodeaction subprocess to fail.

Suppress by passing net_vdev_netvsc,ignore=1 in the valid vdev
test cases. The scan callback sees it already in devargs and
skips injection; the probe honours ignore and does nothing.

Fixes: 0d684a783088 ("app/test: fix and separate --vdev unit test")
Cc: stable@dpdk.org

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
---
 app/test/test_eal_flags.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/app/test/test_eal_flags.c b/app/test/test_eal_flags.c
index bebddf1c98..1c0d8e040a 100644
--- a/app/test/test_eal_flags.c
+++ b/app/test/test_eal_flags.c
@@ -426,13 +426,16 @@ test_invalid_vdev_flag(void)
 
 	/* Test with valid vdev option */
 	const char *vdevval1[] = {prgname, prefix, no_huge, eal_debug_logs,
-				bus_debug_logs,	vdev, "net_ring0"};
+				bus_debug_logs,	vdev, "net_ring0",
+				vdev, "net_vdev_netvsc,ignore=1"};
 
 	const char *vdevval2[] = {prgname, prefix, no_huge, eal_debug_logs,
-				bus_debug_logs, vdev, "net_ring0,args=test"};
+				bus_debug_logs, vdev, "net_ring0,args=test",
+				vdev, "net_vdev_netvsc,ignore=1"};
 
 	const char *vdevval3[] = {prgname, prefix, no_huge, eal_debug_logs,
-				bus_debug_logs, vdev, "net_ring0,nodeaction=r1:0:CREATE"};
+				bus_debug_logs, vdev, "net_ring0,nodeaction=r1:0:CREATE",
+				vdev, "net_vdev_netvsc,ignore=1"};
 
 	if (launch_proc(vdevinval) == 0) {
 		printf("Error (line %d) - process did run ok with invalid vdev parameter\n",
-- 
2.51.0

Re: [RFC] test/eal_flags: suppress vdev_netvsc auto-probe

[Stephen Hemminger]

On Sun, 22 Feb 2026 14:15:42 -0800
Stephen Hemminger <stephen@networkplumber.org> wrote:

> On Hyper-V systems (including Azure-hosted CI), the vdev_netvsc
> driver auto-injects itself during vdev bus scan. This interferes
> with the vdev flag test causing the nodeaction subprocess to fail.
> 
> Suppress by passing net_vdev_netvsc,ignore=1 in the valid vdev
> test cases. The scan callback sees it already in devargs and
> skips injection; the probe honours ignore and does nothing.
> 
> Fixes: 0d684a783088 ("app/test: fix and separate --vdev unit test")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>

Self NAK

This is not the underlying cause of the test failures.
The issue is likely parallel test invocation overlapping on the file prefix
which is already addressed by another pending patch series.

That said, something should be done about vdev_netvsc driver and github
interactions.

Re: [RFC] test/eal_flags: suppress vdev_netvsc auto-probe

[Bruce Richardson]

On Sun, Feb 22, 2026 at 06:49:44PM -0800, Stephen Hemminger wrote:
> On Sun, 22 Feb 2026 14:15:42 -0800
> Stephen Hemminger <stephen@networkplumber.org> wrote:
> 
> > On Hyper-V systems (including Azure-hosted CI), the vdev_netvsc
> > driver auto-injects itself during vdev bus scan. This interferes
> > with the vdev flag test causing the nodeaction subprocess to fail.
> > 
> > Suppress by passing net_vdev_netvsc,ignore=1 in the valid vdev
> > test cases. The scan callback sees it already in devargs and
> > skips injection; the probe honours ignore and does nothing.
> > 
> > Fixes: 0d684a783088 ("app/test: fix and separate --vdev unit test")
> > Cc: stable@dpdk.org
> > 
> > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
> 
> Self NAK
> 
> This is not the underlying cause of the test failures.
> The issue is likely parallel test invocation overlapping on the file prefix
> which is already addressed by another pending patch series.
> 
> That said, something should be done about vdev_netvsc driver and github
> interactions.

I don't think it's parallel invocations - we don't run our unit tests in
parallel in github actions. I'd also note that the current issues started
appearing when we patched in support to pass through the drivers path to
the secondary process calls so that they could match the primary instance.
Therefore, I still suspect something in the pci or other bus scanning to be
a problem. Do we need to disable the netvsc as in your patch and also pass
in --no-pci to the test cases?

/Bruce

Re: [RFC] test/eal_flags: suppress vdev_netvsc auto-probe

[Stephen Hemminger]

On Mon, 23 Feb 2026 17:00:11 +0000
Bruce Richardson <bruce.richardson@intel.com> wrote:

> On Sun, Feb 22, 2026 at 06:49:44PM -0800, Stephen Hemminger wrote:
> > On Sun, 22 Feb 2026 14:15:42 -0800
> > Stephen Hemminger <stephen@networkplumber.org> wrote:
> >   
> > > On Hyper-V systems (including Azure-hosted CI), the vdev_netvsc
> > > driver auto-injects itself during vdev bus scan. This interferes
> > > with the vdev flag test causing the nodeaction subprocess to fail.
> > > 
> > > Suppress by passing net_vdev_netvsc,ignore=1 in the valid vdev
> > > test cases. The scan callback sees it already in devargs and
> > > skips injection; the probe honours ignore and does nothing.
> > > 
> > > Fixes: 0d684a783088 ("app/test: fix and separate --vdev unit test")
> > > Cc: stable@dpdk.org
> > > 
> > > Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>  
> > 
> > Self NAK
> > 
> > This is not the underlying cause of the test failures.
> > The issue is likely parallel test invocation overlapping on the file prefix
> > which is already addressed by another pending patch series.
> > 
> > That said, something should be done about vdev_netvsc driver and github
> > interactions.  
> 
> I don't think it's parallel invocations - we don't run our unit tests in
> parallel in github actions. I'd also note that the current issues started
> appearing when we patched in support to pass through the drivers path to
> the secondary process calls so that they could match the primary instance.
> Therefore, I still suspect something in the pci or other bus scanning to be
> a problem. Do we need to disable the netvsc as in your patch and also pass
> in --no-pci to the test cases?
> 
> /Bruce

I am thinking that vdev_netvsc driver hack needs to have a way
of detecting whether it is being run in a test. Maybe getenv("DPDK_TEST")
or some other github detection.