2019-05-27 07:01:13

by Mao Han

[permalink] [raw]
Subject: [PATCH V2 0/5] csky: Add pmu hardware sampling support

This patch set add hardware sampling support for csky-pmu, and
also add some properties to pmu node definition. perf can record
on hardware event with this patch applied.

Changes since v1:
- do not enable/disable irq at start/stop(may lose irq)
- do not update hpcr when event type is invalid

CC: Guo Ren <[email protected]>
CC: [email protected]

Guo Ren (1):
csky: Fixup some error count in 810 & 860.

Mao Han (4):
csky: Init pmu as a device
csky: Add reg-io-width property for csky pmu
csky: Add pmu interrupt support
dt-bindings: csky: Add csky PMU bindings

Documentation/devicetree/bindings/csky/pmu.txt | 38 +++
arch/csky/kernel/perf_event.c | 407 +++++++++++++++++++++++--
2 files changed, 424 insertions(+), 21 deletions(-)
create mode 100644 Documentation/devicetree/bindings/csky/pmu.txt

--
2.7.4


2019-05-27 07:17:20

by Mao Han

[permalink] [raw]
Subject: [PATCH V2 2/5] csky: Add reg-io-width property for csky pmu

csky pmu counter may have different io width. When the counter is smaller
then 64 bits and counter value is smaller than the old value, it will
result to a extremely large delta value. So the sampled value should be
extend to 64 bits to avoid this, the extension bits base on the
reg-io-width property from dts.

Signed-off-by: Mao Han <[email protected]>
CC: Guo Ren <[email protected]>
CC: [email protected]
---
arch/csky/kernel/perf_event.c | 16 +++++++++++++++-
1 file changed, 15 insertions(+), 1 deletion(-)

diff --git a/arch/csky/kernel/perf_event.c b/arch/csky/kernel/perf_event.c
index c022acc..f1b3cdf 100644
--- a/arch/csky/kernel/perf_event.c
+++ b/arch/csky/kernel/perf_event.c
@@ -18,6 +18,7 @@ static void (*hw_raw_write_mapping[CSKY_PMU_MAX_EVENTS])(uint64_t val);

struct csky_pmu_t {
struct pmu pmu;
+ uint32_t sign_extend;
uint32_t hpcr;
} csky_pmu;

@@ -806,7 +807,13 @@ static void csky_perf_event_update(struct perf_event *event,
struct hw_perf_event *hwc)
{
uint64_t prev_raw_count = local64_read(&hwc->prev_count);
- uint64_t new_raw_count = hw_raw_read_mapping[hwc->idx]();
+ /*
+ * Extend count value to 64bit, otherwise delta calculation would
+ * be incorrect when overflow occurs.
+ */
+ uint64_t new_raw_count = ((int64_t)hw_raw_read_mapping[hwc->idx]()
+ << csky_pmu.sign_extend)
+ >> csky_pmu.sign_extend;
int64_t delta = new_raw_count - prev_raw_count;

/*
@@ -1037,6 +1044,7 @@ int csky_pmu_device_probe(struct platform_device *pdev,
const struct of_device_id *of_id;
csky_pmu_init init_fn;
struct device_node *node = pdev->dev.of_node;
+ int cnt_width;
int ret = -ENODEV;

of_id = of_match_node(of_table, pdev->dev.of_node);
@@ -1045,6 +1053,12 @@ int csky_pmu_device_probe(struct platform_device *pdev,
ret = init_fn(&csky_pmu);
}

+ if (!of_property_read_u32(node, "reg-io-width", &cnt_width)) {
+ csky_pmu.sign_extend = 64 - cnt_width;
+ } else {
+ csky_pmu.sign_extend = 16;
+ }
+
if (ret) {
pr_notice("[perf] failed to probe PMU!\n");
return ret;
--
2.7.4