[MODERATED] Re: [PATCH 2/2] more sampling fun 2

[mark gross <mgross@linux.intel.com>]

On Thu, Feb 20, 2020 at 07:06:46PM +0000, speck for Ben Hutchings wrote:
> On Thu, 2020-01-16 at 14:16 -0800, speck for mark gross wrote:
> [...]
> > --- a/arch/x86/kernel/cpu/bugs.c
> > +++ b/arch/x86/kernel/cpu/bugs.c
> [...]
> > +void srbds_configure_mitigation(void)
> > +{
> > +	u64 mcu_ctrl;
> > +
> > +	if (!boot_cpu_has_bug(X86_BUG_SRBDS) && !boot_cpu_has_bug(X86_BUG_SRBDS_TSX))
> > +		return;
> > +
> > +	if (!boot_cpu_has(X86_FEATURE_SRBDS_CTRL))
> > +		return;
> > +
> > +	rdmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
> > +	if (srbds_mitigation == SRBDS_MITIGATION_FULL)
> > +		mcu_ctrl &= ~SRBDS_MITG_DIS;
> > +	else if (srbds_mitigation == SRBDS_MITIGATION_OFF)
> > +		mcu_ctrl |= SRBDS_MITG_DIS;
> > +
> > +	if (boot_cpu_has_bug(X86_BUG_SRBDS_TSX) && !boot_cpu_has(X86_FEATURE_RTM))
> > +		mcu_ctrl |= SRBDS_MITG_DIS;
> 
> In this case we will incorrectly report "Mitigation: bus lock when
> using RDRAND or RDSEED" whereas the actual mitigation is that TSX is
> disabled.
I am not calling disabling TSX a mitigation in this case.  If TSX is disabled
then you are not vulnerable and the mitigation can be disabled..

> 
> > +	wrmsrl(MSR_IA32_MCU_OPT_CTRL, mcu_ctrl);
> > +}
> [...]
> > --- a/arch/x86/kernel/cpu/common.c
> > +++ b/arch/x86/kernel/cpu/common.c
> [...] 
> > @@ -1042,6 +1047,19 @@ static const struct x86_cpu_id_ext cpu_vuln_whitelist[] __initconst = {
> >  
> >  	VULNWL_INTEL(CORE_YONAH,		NO_SSB),
> >  
> > +	VULNWL_INTEL(IVYBRIDGE,		SRBDS), /*06_3A*/
> > +	VULNWL_INTEL(HASWELL,		SRBDS), /*06_3C*/
> > +	VULNWL_INTEL(HASWELL_L,		SRBDS), /*06_45*/
> > +	VULNWL_INTEL(HASWELL_G,		SRBDS), /*06_46*/
> > +	VULNWL_INTEL(BROADWELL_G,	SRBDS), /*06_47*/
> > +	VULNWL_INTEL(BROADWELL,		SRBDS), /*06_3D*/
> > +	VULNWL_INTEL(SKYLAKE_L,		SRBDS), /*06_4E*/
> > +	VULNWL_INTEL(SKYLAKE,		SRBDS), /*06_5E*/
> > +	VULNWL_INTEL_STEPPING(KABYLAKE_L, (BIT(0xA)-1),		SRBDS), /*06_8E steppings <=A*/
> 
> But this matches steppings 0-9.
well, with the bitmask its zero based indexing of the stepping.  I'll check
with others to double check my assumption.

> 
> > +	VULNWL_INTEL_STEPPING(KABYLAKE_L, BIT(0xB)|BIT(0xC),	SRBDS_TSX), /*06_8E stepping = 0xB if TSX enabled*/
> > +	VULNWL_INTEL_STEPPING(KABYLAKE, (BIT(0xB)-1),		SRBDS), /*06_9E steppings <=B*/
> 
> And this matches steppings 0-A.
> 
> > +	VULNWL_INTEL_STEPPING(KABYLAKE, BIT(0xC)|BIT(0xD),	SRBDS_TSX), /*06_9E stepping = 0xC if TSX enabled*/
> [...]
> 
> You should write the bit masks using GENMASK() instead of BIT().
Ok I'll swith to GENMASK on the next version.

--mark