[RFC V1 10/11] perf: Expand perf_branch_entry.type

Anshuman Khandual anshuman.khandual at arm.com
Thu Jan 27 20:14:22 PST 2022



On 1/25/22 10:28 PM, James Clark wrote:
> 
> On 24/01/2022 04:30, Anshuman Khandual wrote:
>> Current perf_branch_entry.type is a 4 bits field just enough to accommodate
>> 16 generic branch types. This is insufficient to accommodate platforms like
>> arm64 which has much more branch types. Lets just expands this field into a
>> 6 bits one, which can now hold 64 generic branch types. This also adds more
>> generic branch types and updates the BRBE driver as required.
>>
>> Cc: Peter Zijlstra <peterz at infradead.org>
>> Cc: Ingo Molnar <mingo at redhat.com>
>> Cc: Arnaldo Carvalho de Melo <acme at kernel.org>
>> Cc: Mark Rutland <mark.rutland at arm.com>
>> Cc: Alexander Shishkin <alexander.shishkin at linux.intel.com>
>> Cc: Jiri Olsa <jolsa at redhat.com>
>> Cc: Namhyung Kim <namhyung at kernel.org>
>> Cc: Will Deacon <will at kernel.org>
>> Cc: linux-arm-kernel at lists.infradead.org
>> Cc: linux-perf-users at vger.kernel.org
>> Cc: linux-kernel at vger.kernel.org
>> Signed-off-by: Anshuman Khandual <anshuman.khandual at arm.com>
>> ---
>>  drivers/perf/arm_pmu_brbe.c           |  7 ++++++-
>>  include/uapi/linux/perf_event.h       | 10 ++++++++--
>>  tools/include/uapi/linux/perf_event.h | 10 ++++++++--
>>  tools/perf/util/branch.c              |  8 +++++++-
>>  4 files changed, 29 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/perf/arm_pmu_brbe.c b/drivers/perf/arm_pmu_brbe.c
>> index 8d27ad868359..7cd1208c6c58 100644
>> --- a/drivers/perf/arm_pmu_brbe.c
>> +++ b/drivers/perf/arm_pmu_brbe.c
>> @@ -253,12 +253,17 @@ static int brbe_fetch_perf_type(u64 brbinf)
>>  	case BRBINF_TYPE_DEBUG_EXIT:
>>  		return PERF_BR_DEBUG_EXIT;
>>  	case BRBINF_TYPE_SERROR:
>> +		return PERF_BR_SERROR;
>>  	case BRBINF_TYPE_INST_DEBUG:
>> +		return PERF_BR_DEBUG_INST;
>>  	case BRBINF_TYPE_DATA_DEBUG:
>> +		return PERF_BR_DEBUG_DATA;
>>  	case BRBINF_TYPE_ALGN_FAULT:
>> +		return PERF_BR_FAULT_ALGN;
>>  	case BRBINF_TYPE_INST_FAULT:
>> +		return PERF_BR_FAULT_INST;
>>  	case BRBINF_TYPE_DATA_FAULT:
>> -		return PERF_BR_UNKNOWN;
>> +		return PERF_BR_FAULT_DATA;
>>  	default:
>>  		pr_warn("unknown branch type captured\n");
>>  		return PERF_BR_UNKNOWN;
>> diff --git a/include/uapi/linux/perf_event.h b/include/uapi/linux/perf_event.h
>> index b91d0f575d0c..361fdc6b87a0 100644
>> --- a/include/uapi/linux/perf_event.h
>> +++ b/include/uapi/linux/perf_event.h
>> @@ -256,6 +256,12 @@ enum {
>>  	PERF_BR_FIQ		= 13,	/* fiq */
>>  	PERF_BR_DEBUG_HALT	= 14,	/* debug halt */
>>  	PERF_BR_DEBUG_EXIT	= 15,	/* debug exit */
>> +	PERF_BR_DEBUG_INST	= 16,	/* instruciton debug */
>> +	PERF_BR_DEBUG_DATA	= 17,	/* data debug */
>> +	PERF_BR_FAULT_ALGN	= 18,	/* alignment fault */
>> +	PERF_BR_FAULT_DATA	= 19,	/* data fault */
>> +	PERF_BR_FAULT_INST	= 20,	/* instruction fault */
>> +	PERF_BR_SERROR		= 21,	/* system error */
>>  	PERF_BR_MAX,
>>  };
>>  
>> @@ -1370,8 +1376,8 @@ struct perf_branch_entry {
>>  		in_tx:1,    /* in transaction */
>>  		abort:1,    /* transaction abort */
>>  		cycles:16,  /* cycle count to last branch */
>> -		type:4,     /* branch type */
>> -		reserved:40;
>> +		type:6,     /* branch type */
>> +		reserved:38;
>>  };
>>  
>>  union perf_sample_weight {
>> diff --git a/tools/include/uapi/linux/perf_event.h b/tools/include/uapi/linux/perf_event.h
>> index 1882054e8684..9a82b8aaed93 100644
>> --- a/tools/include/uapi/linux/perf_event.h
>> +++ b/tools/include/uapi/linux/perf_event.h
>> @@ -256,6 +256,12 @@ enum {
>>  	PERF_BR_FIQ		= 13,	/* fiq */
>>  	PERF_BR_DEBUG_HALT	= 14,	/* debug halt */
>>  	PERF_BR_DEBUG_EXIT	= 15,	/* debug exit */
>> +	PERF_BR_DEBUG_INST	= 16,	/* instruciton debug */
>> +	PERF_BR_DEBUG_DATA	= 17,	/* data debug */
>> +	PERF_BR_FAULT_ALGN	= 18,	/* alignment fault */
>> +	PERF_BR_FAULT_DATA	= 19,	/* data fault */
>> +	PERF_BR_FAULT_INST	= 20,	/* instruction fault */
>> +	PERF_BR_SERROR		= 21,	/* system error */
>>  	PERF_BR_MAX,
>>  };
>>  
>> @@ -1370,8 +1376,8 @@ struct perf_branch_entry {
>>  		in_tx:1,    /* in transaction */
>>  		abort:1,    /* transaction abort */
>>  		cycles:16,  /* cycle count to last branch */
>> -		type:4,     /* branch type */
>> -		reserved:40;
>> +		type:6,     /* branch type */
>> +		reserved:38;
>>  };
> There's another copy of this struct in branch.h that is used to access the same data in
> perf which also needs updating:
> 
> 	struct branch_flags {
> 		union {
> 			u64 value;
> 			struct {
> 				u64 mispred:1;
> 				u64 predicted:1;
> 				u64 in_tx:1;
> 				u64 abort:1;
> 				u64 cycles:16;
> 				u64 type:4;
> 				u64 reserved:40;
> 			};
> 		};
> 	};

Sure, thanks for the catch. Will fix it.

> 
> It's never assigned directly but there is some casting stuff going on in
> evsel__parse_sample() and it eventually ends up being used to access branch
> records. Same applies to the privilege data change.
> 

Okay, will do the necessary changes in the privilege data patch.



More information about the linux-arm-kernel mailing list