2022-01-13 17:54:43

by German Gomez

Subject: [PATCH 0/2] perf: arm_spe: make the PMSCR CX bit[3] consistent across the session

The value of the CX bit of the PMSCR register is not consistent across
a perf session. There is an example in [1/2] to reproduce the issue.

This cset applies a small correction to fix the consistency issue.

- [PATCH 1/2] Makes the CX bit consistent by caching the value during
the initialization of the SPE PMU event.
- [PATCH 2/2] Allows CONTEXT packets when profiling in CPU mode.

German Gomez (2):
perf: arm_spe: make the PMSCR CX bit[3] consistent across the session
perf: arm_spe: Enable CONTEXT packets if profiling in CPU mode

drivers/perf/arm_spe_pmu.c | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)

--
2.25.1

2022-01-13 17:54:46

by German Gomez

Subject: [PATCH 1/2] perf: arm_spe: make the PMSCR CX bit[3] consistent across the session

The ARM SPE driver will seenablet the CX bit in the PMSCR register if it
detects that the profiler (perf tool) has enough capabilities during the
initialization of the PMU event in the "arm_spe_pmu_event_init" callback:

reg = arm_spe_event_to_pmscr(event);
if (!perfmon_capable() &&
(reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
BIT(SYS_PMSCR_EL1_CX_SHIFT) |
BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
return -EACCES;

This implies that the CX bit should remain consistent across the entire
duration of the session. However, the value of the bit is later being
changed (from 0 to 1 and vice versa) in the "arm_spe_pmu_start" callback

Consider this example:

1. Run a process in the background with privileges in CPU0

$ taskset --cpu-list 0 sudo dd if=/dev/random of=/dev/null &
[3] 3806

2. Begin a perf session without privileges (we shouldn't see CONTEXT packets)

$ perf record -e arm_spe_0// -C0 -- sleep 1
$ perf report -D | grep CONTEXT
. 0000000e: 65 df 0e 00 00 CONTEXT 0xedf el2
. 0000004e: 65 df 0e 00 00 CONTEXT 0xedf el2
. 0000008e: 65 df 0e 00 00 CONTEXT 0xedf el2
[...]

One way to fix this is by caching the value of the CX bit during the
initialization of the PMU event, so that it remains consistent during
the session.

Signed-off-by: German Gomez <[email protected]>
---
drivers/perf/arm_spe_pmu.c | 6 ++++--
1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c
index d44bcc29d..8515bf85c 100644
--- a/drivers/perf/arm_spe_pmu.c
+++ b/drivers/perf/arm_spe_pmu.c
@@ -57,6 +57,7 @@ struct arm_spe_pmu {
u16 pmsver;
u16 min_period;
u16 counter_sz;
+ bool pmscr_cx;

#define SPE_PMU_FEAT_FILT_EVT (1UL << 0)
#define SPE_PMU_FEAT_FILT_TYP (1UL << 1)
@@ -260,6 +261,7 @@ static const struct attribute_group *arm_spe_pmu_attr_groups[] = {
static u64 arm_spe_event_to_pmscr(struct perf_event *event)
{
struct perf_event_attr *attr = &event->attr;
+ struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu);
u64 reg = 0;

reg |= ATTR_CFG_GET_FLD(attr, ts_enable) << SYS_PMSCR_EL1_TS_SHIFT;
@@ -272,7 +274,7 @@ static u64 arm_spe_event_to_pmscr(struct perf_event *event)
if (!attr->exclude_kernel)
reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);

- if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
+ if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && spe_pmu->pmscr_cx)
reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);

return reg;
@@ -709,10 +711,10 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
!(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT))
return -EOPNOTSUPP;

+ spe_pmu->pmscr_cx = perfmon_capable();
reg = arm_spe_event_to_pmscr(event);
if (!perfmon_capable() &&
(reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
- BIT(SYS_PMSCR_EL1_CX_SHIFT) |
BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
return -EACCES;

--
2.25.1

2022-01-13 17:54:50

by German Gomez

Subject: [PATCH 2/2] perf: arm_spe: Enable CONTEXT packets if profiling in CPU mode

Enable CONTEXT packets in SPE traces if the profiler runs in CPU mode.

Signed-off-by: German Gomez <[email protected]>
---
drivers/perf/arm_spe_pmu.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c
index 8515bf85c..7d9a7fa4f 100644
--- a/drivers/perf/arm_spe_pmu.c
+++ b/drivers/perf/arm_spe_pmu.c
@@ -711,7 +711,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
!(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT))
return -EOPNOTSUPP;

- spe_pmu->pmscr_cx = perfmon_capable();
+ spe_pmu->pmscr_cx = perfmon_capable() || (event->cpu != -1);
reg = arm_spe_event_to_pmscr(event);
if (!perfmon_capable() &&
(reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
--
2.25.1

2022-01-14 09:32:09

by German Gomez

Subject: Re: [PATCH 0/2] perf: arm_spe: make the PMSCR CX bit[3] consistent across the session

Apologies everyone, this was a missend.

On 13/01/2022 17:54, German Gomez wrote:
> The value of the CX bit of the PMSCR register is not consistent across
> a perf session. There is an example in [1/2] to reproduce the issue.
>
> This cset applies a small correction to fix the consistency issue.
>
> - [PATCH 1/2] Makes the CX bit consistent by caching the value during
> the initialization of the SPE PMU event.
> - [PATCH 2/2] Allows CONTEXT packets when profiling in CPU mode.
>
> German Gomez (2):
> perf: arm_spe: make the PMSCR CX bit[3] consistent across the session
> perf: arm_spe: Enable CONTEXT packets if profiling in CPU mode
>
> drivers/perf/arm_spe_pmu.c | 10 ++++++++--
> 1 file changed, 8 insertions(+), 2 deletions(-)
>