[PATCH 2/2] perf arm-spe: Parse more SPE fields and store source

German Gomez german.gomez at arm.com
Fri Jan 28 09:20:23 PST 2022


Hi Ali,

On 25/01/2022 19:20, Ali Saidi wrote:
> Decode more SPE events and op types to allow for processing by perf
> scripts. For example looking for branches which may indicate candidates
> for conversion to a CSEL, store exclusives that are candidates for
> conversion to LSE atomics and record the source information for memory
> ops.
>
> Signed-off-by: Ali Saidi <alisaidi at amazon.com>
> ---
>  .../util/arm-spe-decoder/arm-spe-decoder.c     | 18 ++++++++++++++++++
>  .../util/arm-spe-decoder/arm-spe-decoder.h     |  8 ++++++++
>  2 files changed, 26 insertions(+)
>
> diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
> index 5e390a1a79ab..177bac0f7128 100644
> --- a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
> +++ b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.c
> @@ -191,6 +191,20 @@ static int arm_spe_read_record(struct arm_spe_decoder *decoder)
>  					decoder->record.op = ARM_SPE_ST;
>  				else
>  					decoder->record.op = ARM_SPE_LD;
> +				if (SPE_OP_PKT_IS_LDST_ATOMIC(payload)) {
> +					if (payload & SPE_OP_PKT_AT)
> +						decoder->record.op |= ARM_SPE_LDST_ATOMIC;

In "utils/arm-spe.c" we check "if (record->op == ARM_SPE_LD)" so this
ORing could break some of the generated samples.

> +					if (payload & SPE_OP_PKT_EXCL)
> +						decoder->record.op |= ARM_SPE_LDST_EXCL;
> +					if (payload & SPE_OP_PKT_AR)
> +						decoder->record.op |= ARM_SPE_LDST_ACQREL;
> +				}
> +			} else if (idx == SPE_OP_PKT_HDR_CLASS_BR_ERET) {
> +				decoder->record.op = ARM_SPE_BR;
> +				if (payload & SPE_OP_PKT_COND)
> +					decoder->record.op |= ARM_SPE_BR_COND;
> +				if (SPE_OP_PKT_IS_INDIRECT_BRANCH(payload))
> +					decoder->record.op |= ARM_SPE_BR_IND;
>  			}
>  			break;
>  		case ARM_SPE_EVENTS:
> @@ -218,8 +232,12 @@ static int arm_spe_read_record(struct arm_spe_decoder *decoder)
>  			if (payload & BIT(EV_MISPRED))
>  				decoder->record.type |= ARM_SPE_BRANCH_MISS;
>  
> +			if (payload & BIT(EV_NOT_TAKEN))
> +				decoder->record.type |= ARM_SPE_BR_NOT_TAKEN;
> +
>  			break;
>  		case ARM_SPE_DATA_SOURCE:
> +			decoder->record.source = payload;
>  			break;
>  		case ARM_SPE_BAD:
>  			break;
> diff --git a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
> index 69b31084d6be..113e427afe99 100644
> --- a/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
> +++ b/tools/perf/util/arm-spe-decoder/arm-spe-decoder.h
> @@ -22,11 +22,18 @@ enum arm_spe_sample_type {
>  	ARM_SPE_TLB_MISS	= 1 << 5,
>  	ARM_SPE_BRANCH_MISS	= 1 << 6,
>  	ARM_SPE_REMOTE_ACCESS	= 1 << 7,
> +	ARM_SPE_BR_NOT_TAKEN	= 1 << 8,

Can you rename it to ARM_SPE_BRANCH_NOT_TAKEN for consistency?

>  };
>  
>  enum arm_spe_op_type {
>  	ARM_SPE_LD		= 1 << 0,
>  	ARM_SPE_ST		= 1 << 1,
> +	ARM_SPE_LDST_EXCL	= 1 << 2,
> +	ARM_SPE_LDST_ATOMIC	= 1 << 3,
> +	ARM_SPE_LDST_ACQREL	= 1 << 4,
> +	ARM_SPE_BR		= 1 << 5,
> +	ARM_SPE_BR_COND		= 1 << 6,
> +	ARM_SPE_BR_IND		= 1 << 7,

I'm not sure if we should keep everything in one enum/bitmask. I'm also
looking at adding more of the data from the packets to the record, and
considering refactoring the record structure. I'll share here when I
have something.

Thanks,
German

>  };
>  
>  struct arm_spe_record {
> @@ -40,6 +47,7 @@ struct arm_spe_record {
>  	u64 virt_addr;
>  	u64 phys_addr;
>  	u64 context_id;
> +	u16 source;
>  };
>  
>  struct arm_spe_insn;



More information about the linux-arm-kernel mailing list