[PATCH] mshv: add vmbus dependency

[PATCH] mshv: add vmbus dependency

[Arnd Bergmann]

From: Arnd Bergmann <arnd@arndb.de>

When the vmbus driver is not part of the kernel, the mvhv_root
driver now fails to link:

ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!

Avoid this by adding an explicit Kconfig dependency. Note that
stubbing out the hv_vmbus_exists() based on configuration would
also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.

Fixes: f1a9e67c1138 ("mshv: limit SynIC management to MSHV-owned resources")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
---
 drivers/hv/Kconfig | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/hv/Kconfig b/drivers/hv/Kconfig
index 52af086fdeb2..21193b571a80 100644
--- a/drivers/hv/Kconfig
+++ b/drivers/hv/Kconfig
@@ -75,6 +75,7 @@ config MSHV_ROOT
 	# e.g. When withdrawing memory, the hypervisor gives back 4k pages in
 	# no particular order, making it impossible to reassemble larger pages
 	depends on PAGE_SIZE_4KB
+	depends on HYPERV_VMBUS
 	select EVENTFD
 	select VIRT_XFER_TO_GUEST_WORK
 	select HMM_MIRROR
-- 
2.39.5

RE: [PATCH] mshv: add vmbus dependency

[Michael Kelley]

From: Arnd Bergmann <arnd@kernel.org> Sent: Wednesday, May 20, 2026 12:40 AM
> 
> When the vmbus driver is not part of the kernel, the mvhv_root
> driver now fails to link:
> 
> ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!
> 
> Avoid this by adding an explicit Kconfig dependency. Note that
> stubbing out the hv_vmbus_exists() based on configuration would
> also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.

Conceptually, the MSHV root code should not have a dependency on
VMBus. The "does VMBus exist?" question should handled differently
by setting up a boolean in the core Hyper-V code that defaults to "false".
If the VMBus driver loads, it would set the boolean to "true". MSHV
root code would query the boolean.

Michael

> 
> Fixes: f1a9e67c1138 ("mshv: limit SynIC management to MSHV-owned resources")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
>  drivers/hv/Kconfig | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/hv/Kconfig b/drivers/hv/Kconfig
> index 52af086fdeb2..21193b571a80 100644
> --- a/drivers/hv/Kconfig
> +++ b/drivers/hv/Kconfig
> @@ -75,6 +75,7 @@ config MSHV_ROOT
>  	# e.g. When withdrawing memory, the hypervisor gives back 4k pages in
>  	# no particular order, making it impossible to reassemble larger pages
>  	depends on PAGE_SIZE_4KB
> +	depends on HYPERV_VMBUS
>  	select EVENTFD
>  	select VIRT_XFER_TO_GUEST_WORK
>  	select HMM_MIRROR
> --
> 2.39.5
> 

Re: [PATCH] mshv: add vmbus dependency

[Arnd Bergmann]

On Wed, May 20, 2026, at 15:46, Michael Kelley wrote:
> From: Arnd Bergmann <arnd@kernel.org> Sent: Wednesday, May 20, 2026 12:40 AM
>> 
>> When the vmbus driver is not part of the kernel, the mvhv_root
>> driver now fails to link:
>> 
>> ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!
>> 
>> Avoid this by adding an explicit Kconfig dependency. Note that
>> stubbing out the hv_vmbus_exists() based on configuration would
>> also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.
>
> Conceptually, the MSHV root code should not have a dependency on
> VMBus. The "does VMBus exist?" question should handled differently
> by setting up a boolean in the core Hyper-V code that defaults to "false".
> If the VMBus driver loads, it would set the boolean to "true". MSHV
> root code would query the boolean.

That makes sense to me, but it's outside of my area for a drive-by
build fix. Please treat my patch as a 'Reported-by' then, I expect
this can easily be addressed by Jork or someone else in the
hyperv team.

      Arnd

Re: [PATCH] mshv: add vmbus dependency

[Jork Loeser]

On Wed, 20 May 2026, Arnd Bergmann wrote:

> On Wed, May 20, 2026, at 15:46, Michael Kelley wrote:
>> From: Arnd Bergmann <arnd@kernel.org> Sent: Wednesday, May 20, 2026 12:40 AM
>>>
>>> When the vmbus driver is not part of the kernel, the mvhv_root
>>> driver now fails to link:
>>>
>>> ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!
>>>
>>> Avoid this by adding an explicit Kconfig dependency. Note that
>>> stubbing out the hv_vmbus_exists() based on configuration would
>>> also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.
>>
>> Conceptually, the MSHV root code should not have a dependency on
>> VMBus. The "does VMBus exist?" question should handled differently
>> by setting up a boolean in the core Hyper-V code that defaults to "false".
>> If the VMBus driver loads, it would set the boolean to "true". MSHV
>> root code would query the boolean.
>
> That makes sense to me, but it's outside of my area for a drive-by
> build fix. Please treat my patch as a 'Reported-by' then, I expect
> this can easily be addressed by Jork or someone else in the
> hyperv team.

Apologies for letting this linger - we were looking for the best fix 
internally. While Michael is correct that conceptually the dependency is 
not ideal, I think it's the right fix *for now*.

Yes, this would not allow MSHV root without VMBUS, which is relevant for 
MSHV root only, which misses a few other components. For L1VH it is the 
right fix that also enforces module load order.

Ultimately (and this is planned), we need a SYNIC driver, and the 
hv_vmbus_exists() will vanish for this usecase. It is a hack even now.

Best,
Jork

Re: [PATCH] mshv: add vmbus dependency

[Jork Loeser]

On Wed, 20 May 2026, Arnd Bergmann wrote:

> From: Arnd Bergmann <arnd@arndb.de>
>
> When the vmbus driver is not part of the kernel, the mvhv_root
> driver now fails to link:
>
> ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!
>
> Avoid this by adding an explicit Kconfig dependency. Note that
> stubbing out the hv_vmbus_exists() based on configuration would
> also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.
>
> Fixes: f1a9e67c1138 ("mshv: limit SynIC management to MSHV-owned resources")
> Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> ---
> drivers/hv/Kconfig | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/hv/Kconfig b/drivers/hv/Kconfig
> index 52af086fdeb2..21193b571a80 100644
> --- a/drivers/hv/Kconfig
> +++ b/drivers/hv/Kconfig
> @@ -75,6 +75,7 @@ config MSHV_ROOT
> 	# e.g. When withdrawing memory, the hypervisor gives back 4k pages in
> 	# no particular order, making it impossible to reassemble larger pages
> 	depends on PAGE_SIZE_4KB
> +	depends on HYPERV_VMBUS
> 	select EVENTFD
> 	select VIRT_XFER_TO_GUEST_WORK
> 	select HMM_MIRROR
> -- 
> 2.39.5
>

Yes, this is the right short-term fix. We will need to solve the root case 
(no VMBUS required) with a separate SYNIC driver abstraction.

Reviewed-by: Jork Loeser <jloeser@linux.microsoft.com>

Thank you!
Jork

RE: [PATCH] mshv: add vmbus dependency

[Michael Kelley]

From: Jork Loeser <jloeser@linux.microsoft.com> Sent: Wednesday, May 20, 2026 10:16 AM
> 
> On Wed, 20 May 2026, Arnd Bergmann wrote:
> 
> > From: Arnd Bergmann <arnd@arndb.de>
> >
> > When the vmbus driver is not part of the kernel, the mvhv_root
> > driver now fails to link:
> >
> > ERROR: modpost: "hv_vmbus_exists" [drivers/hv/mshv_root.ko] undefined!
> >
> > Avoid this by adding an explicit Kconfig dependency. Note that
> > stubbing out the hv_vmbus_exists() based on configuration would
> > also work for some cases, but not with MSHV_ROOT=y and HYPERV_VMBUS=m.
> >
> > Fixes: f1a9e67c1138 ("mshv: limit SynIC management to MSHV-owned resources")
> > Signed-off-by: Arnd Bergmann <arnd@arndb.de>
> > ---
> > drivers/hv/Kconfig | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/drivers/hv/Kconfig b/drivers/hv/Kconfig
> > index 52af086fdeb2..21193b571a80 100644
> > --- a/drivers/hv/Kconfig
> > +++ b/drivers/hv/Kconfig
> > @@ -75,6 +75,7 @@ config MSHV_ROOT
> > 	# e.g. When withdrawing memory, the hypervisor gives back 4k pages in
> > 	# no particular order, making it impossible to reassemble larger pages
> > 	depends on PAGE_SIZE_4KB
> > +	depends on HYPERV_VMBUS
> > 	select EVENTFD
> > 	select VIRT_XFER_TO_GUEST_WORK
> > 	select HMM_MIRROR
> > --
> > 2.39.5
> >
> 
> Yes, this is the right short-term fix. We will need to solve the root case
> (no VMBUS required) with a separate SYNIC driver abstraction.
> 
> Reviewed-by: Jork Loeser <jloeser@linux.microsoft.com>
> 

I have what I think is a better way to fix this. It preserves the
ability to build MSHV without VMBus, while also guaranteeing
that VMBus loads first when present. And it is relatively simple --
hv_vmbus_exists() does not need to be moved out of the
VMBus module. Later today I'll post a separate patch for
consideration.

The separate SynIC driver abstraction can still come later
and improve things further.

Michael