Re: [PATCH v2] x86/bugs: Explicitly clear speculative MSR bits

[Pawan Gupta]

On Mon, Nov 28, 2022 at 01:42:26AM +0100, Borislav Petkov wrote:
> On Thu, Nov 24, 2022 at 02:46:50AM -0800, Breno Leitao wrote:
> > Currently x86_spec_ctrl_base is read at boot time, and speculative bits
> > are set if configs are enable, such as MSR[SPEC_CTRL_IBRS] is enabled if
> > CONFIG_CPU_IBRS_ENTRY is configured. These MSR bits are not cleared if
> > the mitigations are disabled.
> >
> > This is a problem when kexec-ing a kernel that has the mitigation
> > disabled, from a kernel that has the mitigation enabled. In this case,
> > the MSR bits are carried forward and not cleared at the boot of the new
> > kernel. This might have some performance degradation that is hard to
> > find.
> >
> > This problem does not happen if the machine is (hard) rebooted, because
> > the bit will be cleared by default.
> >
> > This patch also defines a SPEC_CTRL_MASK macro, so, we can easily track
> > and clear if eventually some new mitigation shows up.
>
> Just remove that sentence - the macro's function is kinda obvious from
> the diff itself.
>
> > Suggested-by: Pawan Gupta <pawan.kumar.gupta@xxxxxxxxxxxxxxx>
> > Signed-off-by: Breno Leitao <leitao@xxxxxxxxxx>
> > ---
> >  arch/x86/include/asm/msr-index.h | 3 +++
> >  arch/x86/kernel/cpu/bugs.c       | 9 ++++++++-
> >  2 files changed, 11 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/x86/include/asm/msr-index.h b/arch/x86/include/asm/msr-index.h
> > index 4a2af82553e4..704f49580ee1 100644
> > --- a/arch/x86/include/asm/msr-index.h
> > +++ b/arch/x86/include/asm/msr-index.h
> > @@ -54,6 +54,9 @@
> >  #define SPEC_CTRL_RRSBA_DIS_S_SHIFT	6	   /* Disable RRSBA behavior */
> >  #define SPEC_CTRL_RRSBA_DIS_S		BIT(SPEC_CTRL_RRSBA_DIS_S_SHIFT)
> >
> > +#define SPEC_CTRL_MASK			(SPEC_CTRL_IBRS | SPEC_CTRL_STIBP | SPEC_CTRL_SSBD \
> > +							| SPEC_CTRL_RRSBA_DIS_S)
>
> Call that SPEC_CTRL_MITIGATIONS_MASK or so to denote what it is - a
> mask of the SPEC_CTRL bits which the kernel toggles when controlling
> mitigations.
>
> A comment above it wouldn't hurt either.
>
> > +
> >  #define MSR_IA32_PRED_CMD		0x00000049 /* Prediction Command */
> >  #define PRED_CMD_IBPB			BIT(0)	   /* Indirect Branch Prediction Barrier */
> >
> > diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
> > index 3e3230cccaa7..88957da1029b 100644
> > --- a/arch/x86/kernel/cpu/bugs.c
> > +++ b/arch/x86/kernel/cpu/bugs.c
> > @@ -137,8 +137,15 @@ void __init check_bugs(void)
> >  	 * have unknown values. AMD64_LS_CFG MSR is cached in the early AMD
> >  	 * init code as it is not enumerated and depends on the family.
> >  	 */
> > -	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL))
> > +	if (boot_cpu_has(X86_FEATURE_MSR_SPEC_CTRL)) {
> >  		rdmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
> > +		/*
> > +		 * Previously running software may have some controls turned ON.
>
> "Previously running software, like kexec for example, ..."
>
> > +		 * Clear them and let kernel decide which controls to use.
>
> "Clear them and let the mitigations setup below set them based on configuration."
>
> > +		 */
> > +		x86_spec_ctrl_base &= ~SPEC_CTRL_MASK;
> > +		wrmsrl(MSR_IA32_SPEC_CTRL, x86_spec_ctrl_base);
>
> So this WRMSR will happen on the BSP only but the SPEC_CTRL MSR is
> per-CPU. As is x86_spec_ctrl_current which tracks it.
>
> So I'd say you don't need that WRMSR here - the proper value will get
> replicated eventually everywhere...

This patch is particularly for the case when user intends to turn off
the mitigations like with mitigations=off. In that case we need the
WRMSR because mitigation selection will simply return without writing to
the MSR on BSP.

As part of AP init x86_spec_ctrl_setup_ap() writes to the MSR even
when the mitigation is turned off, so AP's should have been fine, but I
think there is a subtle bug there as well. For below call:

x86_spec_ctrl_setup_ap(void)
{
	write_spec_ctrl_current(x86_spec_ctrl_base, true);

When x86_spec_ctrl_base is 0 MSR won't be written because of a check in
write_spec_ctrl_current() that doesn't write the MSR when the new value
(0) is same as x86_spec_ctrl_current (initialized to 0).

Below should fix the problem with APs:

---
diff --git a/arch/x86/kernel/cpu/bugs.c b/arch/x86/kernel/cpu/bugs.c
index 3e3230cccaa7..cfc2ed2661fc 100644
--- a/arch/x86/kernel/cpu/bugs.c
+++ b/arch/x86/kernel/cpu/bugs.c
@@ -66,7 +66,7 @@ static DEFINE_MUTEX(spec_ctrl_mutex);
  */
 void write_spec_ctrl_current(u64 val, bool force)
 {
-	if (this_cpu_read(x86_spec_ctrl_current) == val)
+	if (!force && this_cpu_read(x86_spec_ctrl_current) == val)
 		return;
 
 	this_cpu_write(x86_spec_ctrl_current, val);