Re: lapic_suspend/lapic_resume wrong?

[Juergen Gross <jgross@suse.com>]

On 23/11/15 09:01, Ingo Molnar wrote:
> 
> * Juergen Gross <jgross@suse.com> wrote:
> 
>> Hi,
>>
>> while trying to find the reason for a hanging kernel during resume
>> handling I found a strange inconsistency in arch/x86/kernel/apic/apic.c
>> regarding usage of config options.
>>
>> Attached patch addresses this, no test done as I'm not sure whether
>> this is a correct approach. Can you have a look at it, please?
>>
>>
>> Juergen
>>
>> diff --git a/arch/x86/kernel/apic/apic.c b/arch/x86/kernel/apic/apic.c
>> index 2f69e3b..bc06c9d 100644
>> --- a/arch/x86/kernel/apic/apic.c
>> +++ b/arch/x86/kernel/apic/apic.c
>> @@ -2270,6 +2270,7 @@ static struct {
>>  	unsigned int apic_tmict;
>>  	unsigned int apic_tdcr;
>>  	unsigned int apic_thmr;
>> +	unsigned int apic_cmci;
>>  } apic_pm_state;
>>  
>>  static int lapic_suspend(void)
>> @@ -2299,6 +2300,10 @@ static int lapic_suspend(void)
>>  	if (maxlvt >= 5)
>>  		apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
>>  #endif
>> +#ifdef CONFIG_X86_MCE_INTEL
>> +	if (maxlvt >= 6)
>> +		apic_pm_state.apic_cmci = apic_read(APIC_LVTCMCI);
>> +#endif
>>  
>>  	local_irq_save(flags);
>>  	disable_local_APIC();
>> @@ -2355,10 +2360,14 @@ static void lapic_resume(void)
>>  	apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
>>  	apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
>>  	apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
>> -#if defined(CONFIG_X86_MCE_INTEL)
>> +#if defined(CONFIG_X86_THERMAL_VECTOR)
>>  	if (maxlvt >= 5)
>>  		apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
>>  #endif
>> +#if defined(CONFIG_X86_MCE_INTEL)
>> +	if (maxlvt >= 6)
>> +		apic_write(APIC_LVTCMCI, apic_pm_state.apic_cmci);
>> +#endif
>>  	if (maxlvt >= 4)
>>  		apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
>>  	apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
> 
> the x86 bit looks absolutely sensible to me.

Thanks. I'll give it a suspend/resume test and send out a patch.

> Have you checked whether we indeed lose this value over S/R, or is this mostly 
> working fine by accident, due to us executing the CMCI vector initialization via:
> 
>   mce_syscore_resume()->__mcheck_cpu_init_vendor()->mce_intel_feature_init()->intel_init_cmci() 
> 
> on every resume event?

I don't know. I was more concerned what might happen in a kernel
configured with CONFIG_X86_MCE_INTEL but not CONFIG_X86_THERMAL_VECTOR.
I guess on such a kernel the THMR vector could be set to zero causing
some pain (enabled, vector 0?).

> The Xen fix is unrelated, just put into the same patch, right?

Uuh, yes, sorry. Just a relict of testing another fix.


Juergen