Messages in this thread Patch in this message | | | Date | Tue, 21 Nov 2017 09:55:03 -0600 | From | Kim Phillips <> | Subject | [PATCH] perf: arm_spe: add user error messaging |
| |
Make the SPE PMU driver easier to use by elaborating SPE-specific errors at perf record event initialization time.
Example #1: Trouble setting the sampling period:
BEFORE THIS PATCH:
$ ./perf record -e arm_spe_0/ts_enable=1/ -F 30 true Error: The sys_perf_event_open() syscall returned with 22 (Invalid argument) for event (arm_spe_0/ts_enable=1/). /bin/dmesg may provide additional information. No CONFIG_PERF_EVENTS=y kernel support configured?
with nothing in dmesg.
AFTER: like BEFORE above, but this text is in dmesg:
arm_spe_pmu spe-pmu@0: Sample period must be specified with --count
The user takes the corrective action and specifies -c instead of -F, which succeeds.
Example #2: Trouble specifying CPU vs. SPE domains:
BEFORE:
$ ./perf record -e arm_spe_0/ts_enable=1/ -C 3-7 true Error: The arm_spe_0/ts_enable=1/ event is not supported.
AFTER: like BEFORE above, but this text is in dmesg:
arm_spe_pmu spe-pmu@0: Target CPU list includes unsupported CPUs
Example #3: Non-privileged user tries to obtain physical address (pa) data on a non-paranoid system:
BEFORE:
$ ./perf record -e arm_spe_0/ts_enable=1,pa_enable=1/ -C 0 true Error: You may not have permission to collect stats.
Consider tweaking /proc/sys/kernel/perf_event_paranoid, which controls use of the performance events system by unprivileged users (without CAP_SYS_ADMIN).
The current value is -1:
-1: Allow use of (almost) all events by all users Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK >= 0: Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN Disallow raw tracepoint access by users without CAP_SYS_ADMIN >= 1: Disallow CPU event access by users without CAP_SYS_ADMIN >= 2: Disallow kernel profiling by users without CAP_SYS_ADMIN
To make this setting permanent, edit /etc/sysctl.conf too, e.g.:
kernel.perf_event_paranoid = -1
AFTER: like BEFORE above, but this text is in dmesg:
arm_spe_pmu spe-pmu@0: Admin privilege needed for physical addr, time and/or context capture
Example #4: Trying to exclude idle profiling:
BEFORE:
$ sudo ./perf record -e arm_spe_0/ts_enable=1,pa_enable=1/I -c 1 true Error: PMU Hardware doesn't support sampling/overflow-interrupts.
AFTER: like BEFORE above, but this text is in dmesg:
arm_spe_pmu spe-pmu@0: Cannot exclude profiling when idle
Signed-off-by: Kim Phillips <kim.phillips@arm.com> --- I've tried to help solve the PMU error messaging problem various ways from userspace [1], but it hasn't been resolved [2] by the time the SPE driver was pushed upstream last week (now commit d5d9696b0380).
For these reasons, please consider applying this patch for the interim. Thanks!
[1] latest SPE incarnation:
https://www.spinics.net/lists/arm-kernel/msg614362.html
[2] -EINVAL when using CCN but not from the driver itself:
https://www.spinics.net/lists/arm-kernel/msg614025.html
drivers/perf/arm_spe_pmu.c | 33 +++++++++++++++++++++++++-------- 1 file changed, 25 insertions(+), 8 deletions(-)
diff --git a/drivers/perf/arm_spe_pmu.c b/drivers/perf/arm_spe_pmu.c index 8ce262fc2561..6887eb45b401 100644 --- a/drivers/perf/arm_spe_pmu.c +++ b/drivers/perf/arm_spe_pmu.c @@ -659,20 +659,27 @@ static int arm_spe_pmu_event_init(struct perf_event *event) u64 reg; struct perf_event_attr *attr = &event->attr; struct arm_spe_pmu *spe_pmu = to_spe_pmu(event->pmu); + struct device *dev = &spe_pmu->pdev->dev; /* This is, of course, deeply driver-specific */ if (attr->type != event->pmu->type) return -ENOENT; if (event->cpu >= 0 && - !cpumask_test_cpu(event->cpu, &spe_pmu->supported_cpus)) + !cpumask_test_cpu(event->cpu, &spe_pmu->supported_cpus)) { + dev_err_ratelimited(dev, "Target CPU list includes unsupported CPUs\n"); return -ENOENT; + } - if (arm_spe_event_to_pmsevfr(event) & SYS_PMSEVFR_EL1_RES0) + if (arm_spe_event_to_pmsevfr(event) & SYS_PMSEVFR_EL1_RES0) { + dev_err_ratelimited(dev, "Specified event filter is not supported\n"); return -EOPNOTSUPP; + } - if (attr->exclude_idle) + if (attr->exclude_idle) { + dev_err_ratelimited(dev, "Cannot exclude profiling when idle\n"); return -EOPNOTSUPP; + } /* * Feedback-directed frequency throttling doesn't work when we @@ -681,28 +688,38 @@ static int arm_spe_pmu_event_init(struct perf_event *event) * count to reflect that. Instead, just force the user to specify * a sample period. */ - if (attr->freq) + if (attr->freq) { + dev_err_ratelimited(dev, "Sample period must be specified with --count\n"); return -EINVAL; + } reg = arm_spe_event_to_pmsfcr(event); if ((reg & BIT(SYS_PMSFCR_EL1_FE_SHIFT)) && - !(spe_pmu->features & SPE_PMU_FEAT_FILT_EVT)) + !(spe_pmu->features & SPE_PMU_FEAT_FILT_EVT)) { + dev_err_ratelimited(dev, "Unsupported EVT filter\n"); return -EOPNOTSUPP; + } if ((reg & BIT(SYS_PMSFCR_EL1_FT_SHIFT)) && - !(spe_pmu->features & SPE_PMU_FEAT_FILT_TYP)) + !(spe_pmu->features & SPE_PMU_FEAT_FILT_TYP)) { + dev_err_ratelimited(dev, "Unsupported TYP filter\n"); return -EOPNOTSUPP; + } if ((reg & BIT(SYS_PMSFCR_EL1_FL_SHIFT)) && - !(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT)) + !(spe_pmu->features & SPE_PMU_FEAT_FILT_LAT)) { + dev_err_ratelimited(dev, "Unsupported LAT filter\n"); return -EOPNOTSUPP; + } reg = arm_spe_event_to_pmscr(event); if (!capable(CAP_SYS_ADMIN) && (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) | BIT(SYS_PMSCR_EL1_CX_SHIFT) | - BIT(SYS_PMSCR_EL1_PCT_SHIFT)))) + BIT(SYS_PMSCR_EL1_PCT_SHIFT)))) { + dev_err_ratelimited(dev, "Admin privilege needed for physical addr, time and/or context capture\n"); return -EACCES; + } return 0; } -- 2.15.0
| |