[v7,14/21] x86/split_lock: Enable split lock detection by default

A split locked access locks bus and degrades overall memory access
performance. When split lock detection feature is enumerated, enable
the feature by default to find any split lock issue.

Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
---
 arch/x86/kernel/cpu/intel.c | 34 ++++++++++++++++++++++++++++++++++
 1 file changed, 34 insertions(+)

On Wed, 17 Apr 2019, Fenghua Yu wrote:

> A split locked access locks bus and degrades overall memory access
> performance. When split lock detection feature is enumerated, enable
> the feature by default to find any split lock issue.
> 
> Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
> ---
>  arch/x86/kernel/cpu/intel.c | 34 ++++++++++++++++++++++++++++++++++
>  1 file changed, 34 insertions(+)
> 
> diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c
> index 997d683d3c27..6a692d215bef 100644
> --- a/arch/x86/kernel/cpu/intel.c
> +++ b/arch/x86/kernel/cpu/intel.c
> @@ -34,6 +34,8 @@
>  DEFINE_PER_CPU(u64, msr_test_ctl_cache);
>  EXPORT_PER_CPU_SYMBOL_GPL(msr_test_ctl_cache);
>  
> +static bool split_lock_detect_enable;
> +
>  /*
>   * Just in case our CPU detection goes bad, or you have a weird system,
>   * allow a way to override the automatic disabling of MPX.
> @@ -164,6 +166,23 @@ static bool bad_spectre_microcode(struct cpuinfo_x86 *c)
>  	return false;
>  }
>  
> +static void split_lock_update_msr(void *__unused)
> +{
> +	if (split_lock_detect_enable) {
> +		msr_set_bit(MSR_TEST_CTL, TEST_CTL_SPLIT_LOCK_DETECT_SHIFT);
> +		this_cpu_or(msr_test_ctl_cache, TEST_CTL_SPLIT_LOCK_DETECT);
> +	} else {
> +		msr_clear_bit(MSR_TEST_CTL, TEST_CTL_SPLIT_LOCK_DETECT_SHIFT);
> +		this_cpu_and(msr_test_ctl_cache, ~TEST_CTL_SPLIT_LOCK_DETECT);
> +	}

Nothing in this file initializes msr_test_ctl_cache explicitely. Register
caching always requires to read the register and store it in the cache
before doing anything with it. Nothing guarantees that all bits in that MSR
are 0 by default forever.

And once you do that _before_ calling split_lock_update_msr() then you can
spare the RMW in that function.

> +static void init_split_lock_detect(struct cpuinfo_x86 *c)
> +{
> +	if (cpu_has(c, X86_FEATURE_SPLIT_LOCK_DETECT))
> +		split_lock_update_msr(NULL);
> +}
> +
>  static void early_init_intel(struct cpuinfo_x86 *c)
>  {
>  	u64 misc_enable;
> @@ -661,6 +680,8 @@ static void init_intel(struct cpuinfo_x86 *c)
>  {
>  	early_init_intel(c);
>  
> +	init_split_lock_detect(c);

Sigh. Why needs this to be squeezed in the middle of the whole enumeration
stuff? Just because....

init_intel_misc_features() is called at the end and it does also MSR
caching etc. So down there is the right place.

> +
>  	intel_workarounds(c);
>  
>  	/*
> @@ -1032,9 +1053,22 @@ static const struct cpu_dev intel_cpu_dev = {
>  
>  cpu_dev_register(intel_cpu_dev);
>  
> +#undef pr_fmt
> +#define pr_fmt(fmt) "x86/split lock detection: " fmt
> +
> +static void show_split_lock_detection_info(void)
> +{
> +	if (split_lock_detect_enable)
> +		pr_info("enabled\n");
> +	else
> +		pr_info("disabled\n");

This function is truly useful. The else path is never invoked. See the call
site below.

> +}
> +
>  static void __init set_split_lock_detect(void)
>  {
>  	setup_force_cpu_cap(X86_FEATURE_SPLIT_LOCK_DETECT);
> +	split_lock_detect_enable = true;
> +	show_split_lock_detection_info();
>  }

Oh well.

Thanks,

	tglx