[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20220517130822.GA3369@willie-the-truck>
Date: Tue, 17 May 2022 14:08:22 +0100
From: Will Deacon <will@...nel.org>
To: Srinivasarao Pathipati <quic_c_spathi@...cinc.com>
Cc: mark.rutland@....com, peterz@...radead.org, mingo@...hat.com,
acme@...nel.org, alexander.shishkin@...ux.intel.com,
jolsa@...nel.org, namhyung@...nel.org, catalin.marinas@....com,
linux-arm-kernel@...ts.infradead.org,
linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH V4] arm64: perf: Set PMCR.X of PMCR_EL0 during pmu reset
On Mon, May 16, 2022 at 06:25:38PM +0530, Srinivasarao Pathipati wrote:
> Enable exporting of events over PMU event export bus by setting
> PMCR.X of PMCR_EL0 during pmu reset.
>
> As it impacts power consumption make it configurable at bootup
> with kernel arguments and at runtime with sysctl.
>
> Signed-off-by: Srinivasarao Pathipati <quic_c_spathi@...cinc.com>
> ---
> Changes since V3:
> - export bit is now configurable with sysctl
> - enabling export bit on reset instead of retaining
>
> Changes since V2:
> Done below changes as per Will's comments
> - enabling pmcr_x now configurable with kernel parameters and
> by default it is disabled.
>
> Changes since V1:
> - Preserving only PMCR_X bit as per Robin Murphy's comment.
> ---
> Documentation/admin-guide/kernel-parameters.txt | 4 ++++
> Documentation/admin-guide/sysctl/kernel.rst | 8 ++++++++
> arch/arm64/kernel/perf_event.c | 15 +++++++++++++++
> include/linux/perf_event.h | 1 +
> kernel/sysctl.c | 12 ++++++++++++
> 5 files changed, 40 insertions(+)
>
> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
> index de3da15..2139b81 100644
> --- a/Documentation/admin-guide/kernel-parameters.txt
> +++ b/Documentation/admin-guide/kernel-parameters.txt
> @@ -5150,6 +5150,10 @@
> Useful for devices that are detected asynchronously
> (e.g. USB and MMC devices).
>
> + export_pmu_events
> + [KNL] Sets export bit of PMCR_EL0 to enable the exporting of
> + events over PMU event export bus.
Sorry, I should've been clearer ahbout this before: if you add a sysctl,
then you get the kernel cmdline option for free via something like
"sysctl.kernel.export_pmu_events=foo", so I think you can drop this and
the early_param().
> +
> retain_initrd [RAM] Keep initrd memory after extraction
>
> rfkill.default_state=
> diff --git a/Documentation/admin-guide/sysctl/kernel.rst b/Documentation/admin-guide/sysctl/kernel.rst
> index ddccd10..8fbc3a0 100644
> --- a/Documentation/admin-guide/sysctl/kernel.rst
> +++ b/Documentation/admin-guide/sysctl/kernel.rst
> @@ -892,6 +892,14 @@ The default value is 0 (access disabled).
>
> See Documentation/arm64/perf.rst for more information.
>
> +export_pmu_events
> +=================
You should add something like "(arm64 only)" to the title.
> +Controls the export bit(4th bit) of PMCR_EL0 which enables the exporting of
Just say "Controls the PMU export bit (PMCR_EL0.X), which enables ...".
> +events over an IMPLEMENTATION DEFINED PMU event export bus to another device.
> +
> +0: disables exporting of events
> +
> +1: enables exporting of events
Please state that the default value is 0.
> pid_max
> =======
> diff --git a/arch/arm64/kernel/perf_event.c b/arch/arm64/kernel/perf_event.c
> index cb69ff1..271a8c6 100644
> --- a/arch/arm64/kernel/perf_event.c
> +++ b/arch/arm64/kernel/perf_event.c
> @@ -34,6 +34,7 @@
> #define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_ACCESS 0xEC
> #define ARMV8_THUNDER_PERFCTR_L1I_CACHE_PREF_MISS 0xED
>
> +int sysctl_export_pmu_events __read_mostly;
> /*
> * ARMv8 Architectural defined events, not all of these may
> * be supported on any given implementation. Unsupported events will
> @@ -1025,6 +1026,17 @@ static int armv8pmu_filter_match(struct perf_event *event)
> return evtype != ARMV8_PMUV3_PERFCTR_CHAIN;
> }
>
> +static int __init export_pmu_events(char *str)
> +{
> + /* Exporting of events can be enabled at runtime with sysctl or
> + * statically at bootup with kernel parameters.
> + */
> + sysctl_export_pmu_events = 1;
> + return 0;
> +}
> +
> +early_param("export_pmu_events", export_pmu_events);
> +
> static void armv8pmu_reset(void *info)
> {
> struct arm_pmu *cpu_pmu = (struct arm_pmu *)info;
> @@ -1047,6 +1059,9 @@ static void armv8pmu_reset(void *info)
> if (armv8pmu_has_long_event(cpu_pmu))
> pmcr |= ARMV8_PMU_PMCR_LP;
>
> + if (sysctl_export_pmu_events)
> + pmcr |= ARMV8_PMU_PMCR_X;
> +
> armv8pmu_pmcr_write(pmcr);
Hmm, I think this reset path only runs when initialising/onlining a CPU,
so it's not a great user interface where the sysctl is concerned. It's
probably better to hook armv8pmu_start() for this.
> }
>
> diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
> index da75956..7790328 100644
> --- a/include/linux/perf_event.h
> +++ b/include/linux/perf_event.h
> @@ -1311,6 +1311,7 @@ extern void put_callchain_entry(int rctx);
>
> extern int sysctl_perf_event_max_stack;
> extern int sysctl_perf_event_max_contexts_per_stack;
> +extern int sysctl_export_pmu_events;
>
> static inline int perf_callchain_store_context(struct perf_callchain_entry_ctx *ctx, u64 ip)
> {
> diff --git a/kernel/sysctl.c b/kernel/sysctl.c
> index e52b6e3..3b751a2e 100644
> --- a/kernel/sysctl.c
> +++ b/kernel/sysctl.c
> @@ -2008,6 +2008,18 @@ static struct ctl_table kern_table[] = {
> .extra2 = SYSCTL_ONE_THOUSAND,
> },
> #endif
> +#ifdef CONFIG_HW_PERF_EVENTS
> + {
> + .procname = "export_pmu_events",
> + .data = &sysctl_export_pmu_events,
> + .maxlen = sizeof(int),
> + .mode = 0644,
> + .proc_handler = proc_dointvec_minmax,
> + .extra1 = SYSCTL_ZERO,
> + .extra2 = SYSCTL_ONE,
> +
> + },
> +#endif
Since this is arm64-specific, it should live in the arm64 code and not
here. See how we already register 'armv8_pmu_sysctl_table' for the ARMv8
PMU.
Will
Powered by blists - more mailing lists