Re: [PATCH 05/10] drivers/perf: Use PERF_PMU_CAP_NO_SAMPLING consistently

From: Robin Murphy
Date: Wed Mar 13 2024 - 08:03:10 EST


On 2024-03-13 11:11 am, James Clark wrote:


On 12/03/2024 17:34, Robin Murphy wrote:
Our system PMUs fundamentally cannot support the current notion of
sampling events, so now that the core capability has been clarified,
apply it consistently and purge yet more boilerplate.

Signed-off-by: Robin Murphy <robin.murphy@xxxxxxx>
---
drivers/perf/alibaba_uncore_drw_pmu.c | 6 +-----
drivers/perf/amlogic/meson_ddr_pmu_core.c | 3 ++-
drivers/perf/arm-cci.c | 3 ++-
drivers/perf/arm-ccn.c | 12 +-----------
drivers/perf/arm-cmn.c | 3 ++-
drivers/perf/arm_cspmu/arm_cspmu.c | 17 ++++-------------
drivers/perf/arm_dmc620_pmu.c | 4 ++--
drivers/perf/arm_dsu_pmu.c | 12 +-----------
drivers/perf/arm_smmuv3_pmu.c | 6 +-----
drivers/perf/cxl_pmu.c | 3 ++-
drivers/perf/dwc_pcie_pmu.c | 5 +----
drivers/perf/fsl_imx8_ddr_perf.c | 3 ++-
drivers/perf/fsl_imx9_ddr_perf.c | 3 ++-
drivers/perf/hisilicon/hisi_pcie_pmu.c | 4 ++--
drivers/perf/hisilicon/hisi_uncore_pmu.c | 3 ++-
drivers/perf/hisilicon/hns3_pmu.c | 4 ++--
drivers/perf/marvell_cn10k_ddr_pmu.c | 6 +-----
drivers/perf/qcom_l2_pmu.c | 7 +------
drivers/perf/qcom_l3_pmu.c | 7 +------
drivers/perf/thunderx2_pmu.c | 4 ++--
drivers/perf/xgene_pmu.c | 4 ++--
21 files changed, 36 insertions(+), 83 deletions(-)

[...]
diff --git a/drivers/perf/arm-ccn.c b/drivers/perf/arm-ccn.c
index ce26bb773a56..4114349e62dd 100644
--- a/drivers/perf/arm-ccn.c
+++ b/drivers/perf/arm-ccn.c
@@ -713,7 +713,6 @@ static void arm_ccn_pmu_event_release(struct perf_event *event)
static int arm_ccn_pmu_event_init(struct perf_event *event)
{
struct arm_ccn *ccn;
- struct hw_perf_event *hw = &event->hw;
u32 node_xp, type, event_id;
int valid;
int i;
@@ -721,16 +720,6 @@ static int arm_ccn_pmu_event_init(struct perf_event *event)
ccn = pmu_to_arm_ccn(event->pmu);
- if (hw->sample_period) {
- dev_dbg(ccn->dev, "Sampling not supported!\n");
- return -EOPNOTSUPP;
- }
-
- if (has_branch_stack(event)) {
- dev_dbg(ccn->dev, "Can't exclude execution levels!\n");
- return -EINVAL;
- }
-

[...]

diff --git a/drivers/perf/arm_dsu_pmu.c b/drivers/perf/arm_dsu_pmu.c
index f5ea5acaf2f3..3424d165795c 100644
--- a/drivers/perf/arm_dsu_pmu.c
+++ b/drivers/perf/arm_dsu_pmu.c
@@ -544,23 +544,12 @@ static int dsu_pmu_event_init(struct perf_event *event)
{
struct dsu_pmu *dsu_pmu = to_dsu_pmu(event->pmu);
- /* We don't support sampling */
- if (is_sampling_event(event)) {
- dev_dbg(dsu_pmu->pmu.dev, "Can't support sampling events\n");
- return -EOPNOTSUPP;
- }
-
/* We cannot support task bound events */
if (event->cpu < 0 || event->attach_state & PERF_ATTACH_TASK) {
dev_dbg(dsu_pmu->pmu.dev, "Can't support per-task counters\n");
return -EINVAL;
}
- if (has_branch_stack(event)) {
- dev_dbg(dsu_pmu->pmu.dev, "Can't support filtering\n");
- return -EINVAL;
- }
-

I'm assuming that this and the other has_branch_stack() check were
removed because branch stacks don't actually do anything unless sampling
is enabled?

It's a small difference that there is now no error message if you ask
for branch stacks, but it wouldn't have done anything anyway? I suppose
this error message was also not applied very consistently across the
different devices.

Right - the rarity of these checks, plus the fact that in both cases here they give a nonsensical debug message that has nothing whatsoever to do with the actual failing condition, seems to make it clear that they aren't realistically useful.

In general I don't see any good reason for a non-sampling event to be picky about the exact type of samples it isn't collecting.

Thanks,
Robin.