Skip to contentSkip to conversation

net/vdev_netvsc: disable auto-detection by default

RFC In review Stephen Hemminger 2 participants 5 messages Feb 24 Network drivers

lore
Share & tools
lore
#0 lore raw

RFC 1/1 net/vdev_netvsc: disable auto-detection by default

The vdev_netvsc driver auto-injects itself on Hyper-V via an RTE_INIT constructor. This interferes with tests on github actions which uses Azure where the driver probes in forked subprocesses and crashes on uninitialized interrupt instances.

Guard the auto-detection behind RTE_LIBRTE_VDEV_NETVSC_AUTO (disabled by default). The driver must now be explicitly requested with --vdev=net_vdev_netvsc.

---
 config/rte_config.h                    |  3 +++
 doc/guides/nics/vdev_netvsc.rst        | 11 ++++++-----
 doc/guides/rel_notes/release_26_03.rst |  8 ++++++++
 drivers/net/vdev_netvsc/vdev_netvsc.c  | 20 ++++++++++++++++++--
 4 files changed, 35 insertions(+), 7 deletions(-)
config/rte_config.h +3 -0
index a2609fa403..e1977081e6 100644
@@ -153,4 +153,7 @@
 /* DLB2 defines */
 // RTE_LIBRTE_PMD_DLB2_QUELL_STATS is not set
 
+/* vdev_netvsc defines */
+// RTE_LIBRTE_VDEV_NETVSC_AUTO is not set
+
 #endif /* _RTE_CONFIG_H_ */
doc/guides/nics/vdev_netvsc.rst +6 -5
index 8dd5981947..4a3049ef42 100644
@@ -65,8 +65,9 @@ a new NetVSC driver will be integrated.
 Runtime Configuration
 ---------------------
 
-This driver is invoked automatically in Hyper-V VM systems unless the user
-invoked it by command line using ``--vdev=net_vdev_netvsc`` EAL option.
+This driver can be invoked automatically if ``RTE_LIBRTE_VDEV_NETVSC_AUTO``
+is defined in the build configuration. Otherwise, it must be enabled
+using the ``--vdev=net_vdev_netvsc`` EAL command line option.
 
 The following device parameters are supported:
 
@@ -87,12 +88,12 @@ The following device parameters are supported:
 
 - ``ignore`` [int]
 
-  If nonzero, ignores the driver running (actually used to disable the
-  auto-detection in Hyper-V VM).
+  If auto-detection is enabled in the build, this parameter disables the
+  auto-detection of network interfaces.
 
 .. note::
 
    Not specifying either ``iface`` or ``mac`` makes this driver attach itself to
    all unrouted NetVSC interfaces found on the system.
    Specifying the device makes this driver attach itself to the device
-   regardless the device routes.
+   regardless of the device routes.
doc/guides/rel_notes/release_26_03.rst +8 -0
index b4499ec066..43d5b89695 100644
@@ -106,6 +106,14 @@ New Features
   Added handling of the key combination Control+L
   to clear the screen before redisplaying the prompt.
 
+* **Updated Microsoft vdev_netvsc driver.**
+
+  * Disabled automatic detection on Hyper-V/Azure by default.
+    The driver must now be explicitly requested with ``--vdev=net_vdev_netvsc``.
+    The previous auto-detection behavior can be restored by building with
+    ``RTE_LIBRTE_VDEV_NETVSC_AUTO`` defined.
+
+
 Removed Items
 -------------
 
drivers/net/vdev_netvsc/vdev_netvsc.c +18 -2
index f4a84783ce..a0d9d2aea9 100644
@@ -42,7 +42,9 @@
 #define VDEV_NETVSC_ARG_IFACE "iface"
 #define VDEV_NETVSC_ARG_MAC "mac"
 #define VDEV_NETVSC_ARG_FORCE "force"
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 #define VDEV_NETVSC_ARG_IGNORE "ignore"
+#endif
 #define VDEV_NETVSC_PROBE_MS 1000
 
 #define NETVSC_CLASS_ID "{f8615163-df3e-46c5-913f-f2d2f965ed0e}"
@@ -653,7 +655,9 @@ vdev_netvsc_vdev_probe(struct rte_vdev_device *dev)
 		VDEV_NETVSC_ARG_IFACE,
 		VDEV_NETVSC_ARG_MAC,
 		VDEV_NETVSC_ARG_FORCE,
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 		VDEV_NETVSC_ARG_IGNORE,
+#endif
 		NULL,
 	};
 	const char *name = rte_vdev_device_name(dev);
@@ -663,7 +667,9 @@ vdev_netvsc_vdev_probe(struct rte_vdev_device *dev)
 	unsigned int specified = 0;
 	unsigned int matched = 0;
 	int force = 0;
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 	int ignore = 0;
+#endif
 	unsigned int i;
 	int ret;
 
@@ -678,14 +684,18 @@ vdev_netvsc_vdev_probe(struct rte_vdev_device *dev)
 
 		if (!strcmp(pair->key, VDEV_NETVSC_ARG_FORCE))
 			force = !!atoi(pair->value);
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 		else if (!strcmp(pair->key, VDEV_NETVSC_ARG_IGNORE))
 			ignore = !!atoi(pair->value);
+#endif
 		else if (!strcmp(pair->key, VDEV_NETVSC_ARG_IFACE) ||
 			 !strcmp(pair->key, VDEV_NETVSC_ARG_MAC))
 			++specified;
 	}
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 	if (ignore)
 		goto ignore;
+#endif
 	if (specified > 1) {
 		DRV_LOG(ERR, "More than one way used to specify the netvsc"
 			" device.");
@@ -713,7 +723,9 @@ vdev_netvsc_vdev_probe(struct rte_vdev_device *dev)
 	}
 error:
 	++vdev_netvsc_ctx_inst;
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 ignore:
+#endif
 	rte_kvargs_free(kvargs);
 	/* Reset alarm if there are device context created */
 	if (vdev_netvsc_ctx_count) {
@@ -765,9 +777,12 @@ RTE_PMD_REGISTER_ALIAS(VDEV_NETVSC_DRIVER, eth_vdev_netvsc);
 RTE_PMD_REGISTER_PARAM_STRING(net_vdev_netvsc,
 			      VDEV_NETVSC_ARG_IFACE "=<string> "
 			      VDEV_NETVSC_ARG_MAC "=<string> "
-			      VDEV_NETVSC_ARG_FORCE "=<int> "
-			      VDEV_NETVSC_ARG_IGNORE "=<int>");
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
+			      VDEV_NETVSC_ARG_IGNORE "=<int> ");
+#endif
+			      VDEV_NETVSC_ARG_FORCE "=<int>");
 
+#ifdef RTE_LIBRTE_VDEV_NETVSC_AUTO
 /** Compare function for vdev find device operation. */
 static int
 vdev_netvsc_cmp_rte_device(const struct rte_device *dev1,
@@ -808,3 +823,4 @@ RTE_INIT(vdev_netvsc_custom_scan_add)
 	if (rte_hypervisor_get() == RTE_HYPERVISOR_HYPERV)
 		rte_vdev_add_custom_scan(vdev_netvsc_scan_callback, NULL);
 }
+#endif /* RTE_LIBRTE_VDEV_NETVSC_AUTO */
Signature · 1 line
2.51.0
#1 lore raw
Quoted 11 lines · Stephen Hemminger

The vdev_netvsc driver auto-injects itself on Hyper-V via an RTE_INIT constructor. This interferes with tests on github actions which uses Azure where the driver probes in forked subprocesses and crashes on uninitialized interrupt instances.

Guard the auto-detection behind RTE_LIBRTE_VDEV_NETVSC_AUTO (disabled by default). The driver must now be explicitly requested with --vdev=net_vdev_netvsc.

The failure in GHA can be avoided with a simpler:
https://patchwork.dpdk.org/project/dpdk/patch/[email protected]/

Signature · 1 line
David Marchand
#2 lore raw
Quoted 9 lines · David Marchand
Quoted 6 lines · Stephen Hemminger

The vdev_netvsc driver auto-injects itself on Hyper-V via an RTE_INIT constructor. This interferes with tests on github actions which uses Azure where the driver probes in forked subprocesses and crashes on uninitialized interrupt instances.

Quoted 8 lines
Quoted 5 lines

Guard the auto-detection behind RTE_LIBRTE_VDEV_NETVSC_AUTO (disabled by default). The driver must now be explicitly requested with --vdev=net_vdev_netvsc.

The failure in GHA can be avoided with a simpler:
https://patchwork.dpdk.org/project/dpdk/patch/[email protected]/

After rerunning with debug, the vdev netvsc is still probed, but is not at fault for the recent failures we saw.

VMBUS_BUS: vmbus_probe_one_driver(): VMBUS device 7c1e52d8-3f86-7c1e-52d8-3f867c1e52d8 on NUMA socket -1
VMBUS_BUS: rte_vmbus_map_device(): Not managed by UIO driver, skipped
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_ring0
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_vdev_netvsc
DPAA2_BUS: fslmc_vfio_close_group: Get fd by name((null)) failed(-19)
DPAA2_BUS: Unable to close devices -19
DPAA_BUS: dpaa_bus_cleanup(): >>
DPAA_BUS: Portal already cleaned
DPAA_BUS: dpaa_bus_cleanup(): Bus cleanup done
Calling recursive action for test_invalid_vdev_flag
Returned from recursive action for test_invalid_vdev_flag with 0 Cleaning up /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance returning 0 Test OK
RTE>>

Signature · 1 line
David Marchand
#3 lore raw
Quoted 31 lines · On Tue, 24 Feb 2026 08:58:45 +0100 David Marchand <[email protected]>…
Quoted 9 lines · David Marchand
Quoted 6 lines · Stephen Hemminger

The vdev_netvsc driver auto-injects itself on Hyper-V via an RTE_INIT constructor. This interferes with tests on github actions which uses Azure where the driver probes in forked subprocesses and crashes on uninitialized interrupt instances.

Quoted 8 lines
Quoted 5 lines

Guard the auto-detection behind RTE_LIBRTE_VDEV_NETVSC_AUTO (disabled by default). The driver must now be explicitly requested with --vdev=net_vdev_netvsc.

The failure in GHA can be avoided with a simpler:
https://patchwork.dpdk.org/project/dpdk/patch/[email protected]/

After rerunning with debug, the vdev netvsc is still probed, but is not at fault for the recent failures we saw.

VMBUS_BUS: vmbus_probe_one_driver(): VMBUS device 7c1e52d8-3f86-7c1e-52d8-3f867c1e52d8 on NUMA socket -1
VMBUS_BUS: rte_vmbus_map_device(): Not managed by UIO driver, skipped
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_ring0
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_vdev_netvsc
DPAA2_BUS: fslmc_vfio_close_group: Get fd by name((null)) failed(-19)
DPAA2_BUS: Unable to close devices -19
DPAA_BUS: dpaa_bus_cleanup(): >>

Quoted 8 lines

DPAA_BUS: Portal already cleaned
DPAA_BUS: dpaa_bus_cleanup(): Bus cleanup done
Calling recursive action for test_invalid_vdev_flag
Returned from recursive action for test_invalid_vdev_flag with 0 Cleaning up /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance returning 0 Test OK
RTE>>

The DPAA_BUS looks like noise.
And also VMBUS.

Maybe we need a --no-bus flag that shuts off all buses?

#4 lore raw
Quoted 44 lines · Stephen Hemminger
Quoted 31 lines · On Tue, 24 Feb 2026 08:58:45 +0100 David Marchand <[email protected]>…
Quoted 9 lines · David Marchand
Quoted 6 lines · Stephen Hemminger

The vdev_netvsc driver auto-injects itself on Hyper-V via an RTE_INIT constructor. This interferes with tests on github actions which uses Azure where the driver probes in forked subprocesses and crashes on uninitialized interrupt instances.

Quoted 8 lines
Quoted 5 lines

Guard the auto-detection behind RTE_LIBRTE_VDEV_NETVSC_AUTO (disabled by default). The driver must now be explicitly requested with --vdev=net_vdev_netvsc.

The failure in GHA can be avoided with a simpler:
https://patchwork.dpdk.org/project/dpdk/patch/[email protected]/

After rerunning with debug, the vdev netvsc is still probed, but is not at fault for the recent failures we saw.

VMBUS_BUS: vmbus_probe_one_driver(): VMBUS device 7c1e52d8-3f86-7c1e-52d8-3f867c1e52d8 on NUMA socket -1
VMBUS_BUS: rte_vmbus_map_device(): Not managed by UIO driver, skipped
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_ring0
VDEV_BUS: vdev_probe_all_drivers(): Search driver to probe device net_vdev_netvsc
DPAA2_BUS: fslmc_vfio_close_group: Get fd by name((null)) failed(-19)
DPAA2_BUS: Unable to close devices -19
DPAA_BUS: dpaa_bus_cleanup(): >>

Quoted 8 lines

DPAA_BUS: Portal already cleaned
DPAA_BUS: dpaa_bus_cleanup(): Bus cleanup done
Calling recursive action for test_invalid_vdev_flag
Returned from recursive action for test_invalid_vdev_flag with 0 Cleaning up /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance /home/runner/work/dpdk/dpdk/build/app/dpdk-test recursive instance returning 0 Test OK
RTE>>

Quoted 4 lines

The DPAA_BUS looks like noise.
And also VMBUS.

Maybe we need a --no-bus flag that shuts off all buses?

I have this patch in store :-) since I also need this for OVS, I'll send it soon.

Signature · 1 line
David Marchand