Re: [PATCH 5/5] platform/x86/intel/ifs: Trace support for array test

["Joseph, Jithu" <jithu.joseph@intel.com>]

Thanks for the review

On 2/6/2023 8:40 AM, Steven Rostedt wrote:
> On Tue, 31 Jan 2023 15:43:02 -0800
> Jithu Joseph <jithu.joseph@intel.com> wrote:
> 
>> diff --git a/include/trace/events/intel_ifs.h b/include/trace/events/intel_ifs.h
>> index d7353024016c..db43df4139a2 100644
>> --- a/include/trace/events/intel_ifs.h
>> +++ b/include/trace/events/intel_ifs.h
>> @@ -35,6 +35,33 @@ TRACE_EVENT(ifs_status,
>>  		__entry->status)
>>  );
>>  
>> +TRACE_EVENT(ifs_array,
>> +
>> +	TP_PROTO(int cpu, union ifs_array activate, union ifs_array status),
>> +
>> +	TP_ARGS(cpu, activate, status),
>> +
>> +	TP_STRUCT__entry(
>> +		__field(	u64,	status	)
>> +		__field(	int,	cpu	)
>> +		__field(	u32,	arrays	)
>> +		__field(	u16,	bank	)
>> +	),
>> +
>> +	TP_fast_assign(
>> +		__entry->cpu	= cpu;
>> +		__entry->arrays	= activate.array_bitmask;
>> +		__entry->bank	= activate.array_bank;
> 
> Regardless of the "bitfield" discussion on the other patches, this part
> is considered a fast path (although if where it is called, then it may
> not be). I would just have:
> 
> 		__field(	u64,	data	)
> 
> 		__entry->data = status.data;

Will modify it as given below (in-line with your suggestion)

	TP_STRUCT__entry(
		__field(	u64,	activate	)
		__field(	u64,	status	)
		__field(	int,	cpu	)
	),

	TP_fast_assign(
		__entry->activate	= activate.data;
		__entry->status	= status.data;
		__entry->cpu	= cpu;
	),

	TP_printk("cpu: %d, array_list: %.8llx, array_bank: %.4llx, status: %.16llx",
		__entry->cpu,
		__entry->activate & 0xffffffff,
		(__entry->activate >> 32) & 0xffff,
		__entry->status)

In general this is not called from a fast path within, i.e rate of triggering the tests is likely to be
"per minute" (Though in certain scenarios of driver retries it can be more frequent than that)


> 
> 		__entry->data >> 32,
> 		(__entry->data >> 16) & 0xffff,
> 
> Or something similar. That is, move the parsing of the bits to the
> output. libtraceevent should still be able to handle this.
> 

Jithu