Re: [PATCH -next] x86: Fix unused variable 'msr_val' warning

[Ingo Molnar]

* Michael Kelley <mikelley@xxxxxxxxxxxxx> wrote:

> From: Ingo Molnar <mingo.kernel.org@xxxxxxxxx> Sent: Monday, March 22, 2021 2:08 PM
> > 
> > * Xu Yihang <xuyihang@xxxxxxxxxx> wrote:
> > 
> > > Fixes the following W=1 kernel build warning(s):
> > > arch/x86/hyperv/hv_spinlock.c:28:16: warning: variable 'msr_val' set but not used [-
> > Wunused-but-set-variable]
> > >   unsigned long msr_val;
> > >
> > > As Hypervisor Top-Level Functional Specification states in chapter 7.5 Virtual Processor
> > Idle Sleep State, "A partition which possesses the AccessGuestIdleMsr privilege (refer to
> > section 4.2.2) may trigger entry into the virtual processor idle sleep state through a read to
> > the hypervisor-defined MSR HV_X64_MSR_GUEST_IDLE". That means only a read is
> > necessary, msr_val is not uesed, so __maybe_unused should be added.
> > >
> > > Reference:
> > >
> > > https://docs.microsoft.com/en-us/virtualization/hyper-v-on-windows/reference/tlfs
> > >
> > > Reported-by: Hulk Robot <hulkci@xxxxxxxxxx>
> > > Signed-off-by: Xu Yihang <xuyihang@xxxxxxxxxx>
> > > ---
> > >  arch/x86/hyperv/hv_spinlock.c | 2 +-
> > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > >
> > > diff --git a/arch/x86/hyperv/hv_spinlock.c b/arch/x86/hyperv/hv_spinlock.c
> > > index f3270c1fc48c..67bc15c7752a 100644
> > > --- a/arch/x86/hyperv/hv_spinlock.c
> > > +++ b/arch/x86/hyperv/hv_spinlock.c
> > > @@ -25,7 +25,7 @@ static void hv_qlock_kick(int cpu)
> > >
> > >  static void hv_qlock_wait(u8 *byte, u8 val)
> > >  {
> > > -	unsigned long msr_val;
> > > +	unsigned long msr_val __maybe_unused;
> > >  	unsigned long flags;
> > 
> > Please don't add new __maybe_unused annotations to the x86 tree -
> > improve the flow instead to help GCC recognize the initialization
> > sequence better.
> > 
> > Thanks,
> > 
> > 	Ingo
> 
> Could you elaborate on the thinking here, or point to some written
> discussion?   I'm just curious.   In this particular case, it's not a problem
> with the flow or gcc detection.  This code really does read an MSR and
> ignore that value that is read, so it's not clear how gcc would ever
> figure out that's OK.

Yeah, so the canonical way to signal that the msr_val isn't used would 
be to rewrite this as:


	if (READ_ONCE(*byte) == val) {
		unsigned long msr_val;

		rdmsrl(HV_X64_MSR_GUEST_IDLE, msr_val);

		(void)msr_val;
	}

(Also see the patch below - untested.)

This makes it abundantly clear that the rdmsr() msr_val return value 
is not 'maybe' unused, but totally intentionally skipped.

Thanks,

	Ingo

 arch/x86/hyperv/hv_spinlock.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/x86/hyperv/hv_spinlock.c b/arch/x86/hyperv/hv_spinlock.c
index f3270c1fc48c..7d948513ed42 100644
--- a/arch/x86/hyperv/hv_spinlock.c
+++ b/arch/x86/hyperv/hv_spinlock.c
@@ -25,7 +25,6 @@ static void hv_qlock_kick(int cpu)
 
 static void hv_qlock_wait(u8 *byte, u8 val)
 {
-	unsigned long msr_val;
 	unsigned long flags;
 
 	if (in_nmi())
@@ -48,8 +47,14 @@ static void hv_qlock_wait(u8 *byte, u8 val)
 	/*
 	 * Only issue the rdmsrl() when the lock state has not changed.
 	 */
-	if (READ_ONCE(*byte) == val)
+	if (READ_ONCE(*byte) == val) {
+		unsigned long msr_val;
+
 		rdmsrl(HV_X64_MSR_GUEST_IDLE, msr_val);
+
+		(void)msr_val;
+	}
+
 	local_irq_restore(flags);
 }