Re: [RFC PATCH v2 09/11] powerpc: Add support to mask perf interrupts

[Madhavan Srinivasan <maddy@linux.vnet.ibm.com>]

On Monday 01 August 2016 10:59 AM, Nicholas Piggin wrote:
> On Mon,  1 Aug 2016 00:36:27 +0530
> Madhavan Srinivasan <maddy@linux.vnet.ibm.com> wrote:
>
>>   /*
>>    * flags for paca->soft_enabled
>>    */
>>   #define IRQ_DISABLE_LEVEL_NONE		0
>>   #define IRQ_DISABLE_LEVEL_LINUX		1
>> +#define IRQ_DISABLE_LEVEL_PMU		2
>> +
>> +#define MASK_IRQ_LEVEL		IRQ_DISABLE_LEVEL_LINUX | IRQ_DISABLE_LEVEL_PMU
>>   
>>   
>>   #endif /* CONFIG_PPC64 */
>> diff --git a/arch/powerpc/kernel/exceptions-64s.S b/arch/powerpc/kernel/exceptions-64s.S
>> index 2c87e82ecbe4..56dc71b82824 100644
>> --- a/arch/powerpc/kernel/exceptions-64s.S
>> +++ b/arch/powerpc/kernel/exceptions-64s.S
>> @@ -256,11 +256,11 @@ hardware_interrupt_pSeries:
>>   hardware_interrupt_hv:
>>   	BEGIN_FTR_SECTION
>>   		_MASKABLE_EXCEPTION_PSERIES(0x502, hardware_interrupt,
>> -					    EXC_HV, SOFTEN_TEST_HV, IRQ_DISABLE_LEVEL_LINUX)
>> +					    EXC_HV, SOFTEN_TEST_HV, MASK_IRQ_LEVEL)
> So what I was expecting is that each exception handler would specify the
> level (or bit, if we use bitmask) at which it gets disabled. The test code
> will then test the exception level with the enable level (or s/level/mask).
>
> The way you have now is each exception handler specifying the bits which
> cause it to be disabled, but I think that's kind of backwards -- the
> disabler knows which interrupts it wants to disable, the exception handler
> does not know what disablers want to disable it :)

Yep. My bad. Implemented backwards.

>
> So to disable PMU and "linux" interrupts for local_t operations, you would
> have:
>
> local_irq_set_mask(IRQ_DISABLE_LEVEL_LINUX|IRQ_DISABLE_LEVEL_PMU)

Yep. Right now I made the exception handler to specify this and
  soft_irq_set_level() set the level. But this a minor change.

>
> And that would disable both handlers that test with IRQ_DISABLE_LEVEL_LINUX
> and IRQ_DISABLE_LEVEL_PMU
>
> Does that make sense? What do you think?

Yes. it make sense.
But would like to get comments for the SOFTEN_TEST changes.

Maddy

>
> Thanks,
> Nick
>