Skip to contentSkip to conversation

Drivers: hv: vmbus: Skip VMBus module cleanup for non-nested root partition

PATCH Superseded Michael Kelley 4 participants 4 messages Aug 5 – Aug 10

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Drivers: hv: vmbus: Skip VMBus module cleanup for non-nested root partition

The VMBus module initialization function, hv_acpi_init(), currently does nothing when running in the root partition and root is not nested in another VM. But the initialization function reports success, so the VMBus module is indeed loaded. VMBus functionality is not actually needed, but the VMBus module must be loaded so that hv_vmbus_exists() can answer correctly. Furthermore, the mshv_root dependency on the VMBus module is needed as described in the commit message for 840b740a35bf ("mshv: Add conditional VMBus dependency").

Loading the VMBus module without actually initializing it causes failures if the module should later be unloaded. The module unload code tries to clean up things that were never initialized, resulting in memory faults and a panic.

Fix this by having VMBus module exit function perform the same check for non-nested root partition, and do nothing in such a case, just like hv_acpi_init().

In the long run, the code that manages the Hyper-V provided SynIC should be refactored to better coordinate the requirements of root partition scenarios and normal VM scenarios, and to hopefully remove the hv_vmbus_exists() dependnecy between mshv_root and VMBus modules. Preventing the current unload failure scenario is an expediency until such a refactoring is done.

Fixes 7e279d78664aa ("Drivers: hv: vmbus: skip VMBus initialization if Linux is root")

---
v1: https://lore.kernel.org/linux-hyperv/[email protected]/

Changes in v2:
* Use a different solution: Allow the VMBus module to load but have the unload function do nothing for non-nested root
* Change the patch Subject and commit message to reflect the new approach

 drivers/hv/vmbus_drv.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/hv/vmbus_drv.c +3 -0
index e19ec73b0187..f837153427f4 100644
@@ -3024,6 +3024,9 @@ static void __exit vmbus_exit(void)
 {
 	int cpu;
 
+	if (hv_root_partition() && !hv_nested)
+		return;
+
 	unregister_syscore(&hv_synic_syscore);
 
 	hv_remove_kexec_handler();
Signature · 1 line
2.25.1
#1 lore raw
Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Backwards teardown in vmbus_exit() causes failure to free IRQ action, leading to reload failure and potential Use-After-Free of unmapped IRQ handler. --

--- Patch [1]: [PATCH v2 1/1] Drivers: hv: vmbus: Skip VMBus module cleanup for non-nested root partition --- commit d186708fbf26f95dce2f00a8fed8b7ac79496439
Author: Michael Kelley <[email protected]>

Drivers: hv: vmbus: Skip VMBus module cleanup for non-nested root partition

This commit updates the VMBus module exit function to skip cleanup when running in a non-nested root partition, matching the behavior of the initialization function. This prevents memory faults and kernel panics caused by attempting to clean up uninitialized resources during module unload.

Quoted 14 lines
drivers/hv/vmbus_drv.c +3 -0
index 6824bd7cb3c4f..88fc6add26860 100644
@@ -3030,6 +3030,9 @@ static void __exit vmbus_exit(void)
 {
 	int cpu;
 
+	if (hv_root_partition() && !hv_nested)
+		return;
+
 	unregister_syscore(&hv_synic_syscore);
 
 	hv_remove_kexec_handler();

[ ... ]

drivers/hv/vmbus_drv.c

Quoted 5 lines
drivers/hv/vmbus_drv.c +0 -0
 	vmbus_disconnect();
 	if (vmbus_irq == -1)
 		hv_remove_vmbus_handler();
 	else
 		free_percpu_irq(vmbus_irq, &vmbus_evt);

[Severity: High]
This is a pre-existing issue, but does calling free_percpu_irq() here abort and leak the irqaction?

It appears the CPUs have not yet disabled the interrupt, as cpuhp_remove_state() is called much later in this function. Because cpuhp_remove_state() is what triggers hv_synic_cleanup() to disable the percpu IRQ on each CPU, the IRQ is still locally enabled here.

This sequence typically trips the structural safety check in __free_percpu_irq() causing the free operation to abort and return NULL. If the IRQ line fires after the module is unloaded, could the kernel execute an unmapped function pointer?

[ ... ]

drivers/hv/vmbus_drv.c

Quoted 3 lines
drivers/hv/vmbus_drv.c +0 -0
 	bus_unregister(&hv_bus);
 
 	cpuhp_remove_state(hyperv_cpuhp_online);

[Severity: High]
This is related to the pre-existing issue above. Is this execution of cpuhp_remove_state() too late in the teardown sequence?

This step triggers the offline callback hv_synic_cleanup() which calls disable_percpu_irq(vmbus_irq). Because it runs after free_percpu_irq(), the IRQ action is permanently leaked and subsequent attempts to reload the module might fail.

Signature · 1 line
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1
Reviewed-by on v2 1/1
#2 lore raw
In reply to Michael Kelley
Quoted 45 lines · Michael Kelley

The VMBus module initialization function, hv_acpi_init(), currently does nothing when running in the root partition and root is not nested in another VM. But the initialization function reports success, so the VMBus module is indeed loaded. VMBus functionality is not actually needed, but the VMBus module must be loaded so that hv_vmbus_exists() can answer correctly. Furthermore, the mshv_root dependency on the VMBus module is needed as described in the commit message for 840b740a35bf ("mshv: Add conditional VMBus dependency").

Loading the VMBus module without actually initializing it causes failures if the module should later be unloaded. The module unload code tries to clean up things that were never initialized, resulting in memory faults and a panic.

Fix this by having VMBus module exit function perform the same check for non-nested root partition, and do nothing in such a case, just like hv_acpi_init().

In the long run, the code that manages the Hyper-V provided SynIC should be refactored to better coordinate the requirements of root partition scenarios and normal VM scenarios, and to hopefully remove the hv_vmbus_exists() dependnecy between mshv_root and VMBus modules. Preventing the current unload failure scenario is an expediency until such a refactoring is done.

Fixes 7e279d78664aa ("Drivers: hv: vmbus: skip VMBus initialization if Linux is root")

---
v1: https://lore.kernel.org/linux-hyperv/[email protected]/

Changes in v2:
* Use a different solution: Allow the VMBus module to load but have the unload function do nothing for non-nested root
* Change the patch Subject and commit message to reflect the new approach

 drivers/hv/vmbus_drv.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/hv/vmbus_drv.c +0 -0
index e19ec73b0187..f837153427f4 100644

drivers/hv/vmbus_drv.c static void __exit vmbus_exit(void) · line 3029

Quoted 6 lines
drivers/hv/vmbus_drv.c +3 -0
@@ -3024,6 +3024,9 @@ static void __exit vmbus_exit(void)
 {
 	int cpu;
 
+	if (hv_root_partition() && !hv_nested)
+		return;
+

Do you think it'd be useful to have a comment here either saying something to the effect of "nothing was initialized, so let's skip the teardown", or alternatively pointing to the paired check in hv_acpi_init()?

Either way,

Quoted 3 lines
drivers/hv/vmbus_drv.c +0 -0
 	unregister_syscore(&hv_synic_syscore);
 
 	hv_remove_kexec_handler();
#3 lore raw

From: Easwar Hariharan <[email protected]> Sent: Wednesday, August 5, 2026 10:28 AM

Quoted 50 lines · Michael Kelley
Quoted 48 lines

The VMBus module initialization function, hv_acpi_init(), currently does nothing when running in the root partition and root is not nested in another VM. But the initialization function reports success, so the VMBus module is indeed loaded. VMBus functionality is not actually needed, but the VMBus module must be loaded so that hv_vmbus_exists() can answer correctly. Furthermore, the mshv_root dependency on the VMBus module is needed as described in the commit message for 840b740a35bf ("mshv: Add conditional VMBus dependency").

Loading the VMBus module without actually initializing it causes failures if the module should later be unloaded. The module unload code tries to clean up things that were never initialized, resulting in memory faults and a panic.

Fix this by having VMBus module exit function perform the same check for non-nested root partition, and do nothing in such a case, just like hv_acpi_init().

In the long run, the code that manages the Hyper-V provided SynIC should be refactored to better coordinate the requirements of root partition scenarios and normal VM scenarios, and to hopefully remove the hv_vmbus_exists() dependnecy between mshv_root and VMBus modules. Preventing the current unload failure scenario is an expediency until such a refactoring is done.

Fixes 7e279d78664aa ("Drivers: hv: vmbus: skip VMBus initialization if Linux is root")

---
v1: https://lore.kernel.org/linux-hyperv/[email protected]/

Changes in v2:
* Use a different solution: Allow the VMBus module to load but have the unload function do nothing for non-nested root
* Change the patch Subject and commit message to reflect the new approach

 drivers/hv/vmbus_drv.c | 3 +++
 1 file changed, 3 insertions(+)
drivers/hv/vmbus_drv.c +0 -0
index e19ec73b0187..f837153427f4 100644
@@ -3024,6 +3024,9 @@ static void __exit vmbus_exit(void)
 {
 	int cpu;
 
Quoted 7 lines

drivers/hv/vmbus_drv.c static void __exit vmbus_exit(void) · line 3029

Quoted 3 lines
drivers/hv/vmbus_drv.c +3 -0
+	if (hv_root_partition() && !hv_nested)
+		return;
+

Do you think it'd be useful to have a comment here either saying something to the effect of "nothing was initialized, so let's skip the teardown", or alternatively pointing to the paired check in hv_acpi_init()?

I generally lean toward adding comments for anything that is non-obvious, so yes. Actually, a comment is really needed in hv_acpi_init() to explain why the code intentionally allows the module to be loaded even though no initialization is done. Then a comment here can refer back to that.

I'll spin a v3.

Michael