RE: [PATCH hyperv-next v4 04/16] arch/x86: mshyperv: Trap on access for some synthetic MSRs

[Michael Kelley]

From: Roman Kisel <romank@xxxxxxxxxxxxxxxxxxx> Sent: Monday, July 14, 2025 3:16 PM
> 
> hv_set_non_nested_msr() has special handling for SINT MSRs
> when a paravisor is present. In addition to updating the MSR on the
> host, the mirror MSR in the paravisor is updated, including with the
> proxy bit. But with Confidential VMBus, the proxy bit must not be
> used, so add a special case to skip it.
> 
> Update the hv_set_non_nested_msr() function as well as
> vmbus_signal_eom() to trap on access for some synthetic MSRs.

This paragraph looks like a leftover from the previous version of
the commit message.  vmbus_signal_eom() is not touched in this
patch.

> 
> Signed-off-by: Roman Kisel <romank@xxxxxxxxxxxxxxxxxxx>
> Reviewed-by: Alok Tiwari <alok.a.tiwari@xxxxxxxxxx>
> ---
>  arch/x86/kernel/cpu/mshyperv.c | 17 +++++++++++++----
>  1 file changed, 13 insertions(+), 4 deletions(-)
> 
> diff --git a/arch/x86/kernel/cpu/mshyperv.c b/arch/x86/kernel/cpu/mshyperv.c
> index 07c60231d0d8..6c5a0a779c02 100644
> --- a/arch/x86/kernel/cpu/mshyperv.c
> +++ b/arch/x86/kernel/cpu/mshyperv.c
> @@ -28,6 +28,7 @@
>  #include <asm/apic.h>
>  #include <asm/timer.h>
>  #include <asm/reboot.h>
> +#include <asm/msr.h>
>  #include <asm/nmi.h>
>  #include <clocksource/hyperv_timer.h>
>  #include <asm/msr.h>
> @@ -79,13 +80,21 @@ EXPORT_SYMBOL_GPL(hv_get_non_nested_msr);
>  void hv_set_non_nested_msr(unsigned int reg, u64 value)
>  {
>  	if (hv_is_synic_msr(reg) && ms_hyperv.paravisor_present) {
> +		/* The hypervisor will get the intercept. */
>  		hv_ivm_msr_write(reg, value);
> 
> -		/* Write proxy bit via wrmsl instruction */
> -		if (hv_is_sint_msr(reg))
> -			wrmsrq(reg, value | 1 << 20);
> +		if (hv_is_sint_msr(reg)) {
> +			/*
> +			 * Write proxy bit in the case of non-confidential VMBus.
> +			 * Using wrmsrq instruction so the following goes to the paravisor.
> +			 */
> +			u32 proxy = vmbus_is_confidential() ? 0 : 1;

The kernel test robot is rightly complaining about vmbus_is_confidential()
being used here. vmbus_is_confidential() comes from vmbus_drv.c, which
is built as a module when CONFIG_HYPERV=m. But this code is always
built-in, and built-in code can't directly refer to functions in a module.
Logically, the layering is wrong because vmbus_is_confidential() is
obviously a VMBus concept, while mshyperv.c only deals with Hyper-V
and knows nothing about VMBus. Your v3 patch series has the same
problem, but evidently the kernel test robot didn't notice the issue,
and I didn't notice it either my v3 review.

The best solution is probably to define a 32-bit "proxy bit" value here
in mshyperv.c. hv_set_non_nested_msr() would always OR in that
value before the call to native_wrmsrq() below. Wherever VMBus code
changes the value of is_confidential, it must also call an EXPORTed
function here in mshyperv.c to set or clear the value of "proxy bit".
Overall, it's a bit messy, but I don't see a better solution.

> +
> +			value |= (proxy << 20);
> +			native_wrmsrq(reg, value);
> +		}
>  	} else {
> -		wrmsrq(reg, value);
> +		native_wrmsrq(reg, value);
>  	}
>  }
>  EXPORT_SYMBOL_GPL(hv_set_non_nested_msr);
> --
> 2.43.0