linux-kernel - Re: [PATCH] drivers/perf: arm_pmu: save/restore cpu cycle counter in cpu_pm_pmu

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20171120123240.ohcdgv2nnoyrc7ov@lakrids.cambridge.arm.com>
Date:   Mon, 20 Nov 2017 12:32:40 +0000
From:   Mark Rutland <mark.rutland@....com>
To:     Jia He <hejianet@...il.com>
Cc:     Will Deacon <will.deacon@....com>,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
        Jia He <jia.he@...-semitech.com>
Subject: Re: [PATCH] drivers/perf: arm_pmu: save/restore cpu cycle counter in
 cpu_pm_pmu_notify

Hi,

On Thu, Nov 16, 2017 at 06:27:28AM +0000, Jia He wrote:
> Sometimes userspace need a high resolution cycle counter by reading
> pmccntr_el0.
> 
> In commit da4e4f18afe0 ("drivers/perf: arm_pmu: implement CPU_PM
> notifier"), it resets all the counters even when the pmcr_el0.E and
> pmcntenset_el0.C are both 1 . That is incorrect.

I appreciate that you may wish to make use of the cycle counter from
userspace, but this is the intended behaviour kernel-side. Direct
userspace counter acceess is not supported.

In power states where context is lost, any perf events are
saved/restored by cpu_pm_pmu_setup(). So we certainly shouldn't be
modifying the counter registers in any other PM code.

We *could* expose counters to userspace on homogeneous systems, so long
as users stuck to the usual perf data page interface. However, this
comes with a number of subtle problems, and no-one's done the work to
enable this.

Even then, perf may modify counters at any point in time, and
monotonicity (and/or presence) of counters is not guaranteed.

> We need to save the registers and counter before CPU_PM_ENTER and
> restore them after CPU_PM_EXIT.
> 
> Fixes: da4e4f18afe0 ("drivers/perf: arm_pmu: implement CPU_PM notifier")

As above, this patch is not a  fix, and is not currently necessary.

Thanks,
Mark.

> Signed-off-by: Jia He <jia.he@...-semitech.com>
> ---
>  drivers/perf/arm_pmu.c | 72 +++++++++++++++++++++++++++++++++++++++++++++-----
>  1 file changed, 66 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
> index 7bc5eee..cf55c91 100644
> --- a/drivers/perf/arm_pmu.c
> +++ b/drivers/perf/arm_pmu.c
> @@ -26,6 +26,12 @@
>  
>  #include <asm/irq_regs.h>
>  
> +#ifdef CONFIG_CPU_PM
> +DEFINE_PER_CPU(u32, saved_pmcr_el0);
> +DEFINE_PER_CPU(u32, saved_pmcntenset_el0);
> +DEFINE_PER_CPU(u64, saved_cycle_cnter);
> +#endif
> +
>  static int
>  armpmu_map_cache_event(const unsigned (*cache_map)
>  				      [PERF_COUNT_HW_CACHE_MAX]
> @@ -719,6 +725,15 @@ static void cpu_pm_pmu_setup(struct arm_pmu *armpmu, unsigned long cmd)
>  	}
>  }
>  
> +static int pmc_cycle_counter_enabled(void)
> +{
> +	if ((read_sysreg(pmcr_el0) & ARMV8_PMU_PMCR_E) &&
> +				read_sysreg(pmcntenset_el0) & 1<<31)
> +		return 1;
> +
> +	return 0;
> +}
> +
>  static int cpu_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
>  			     void *v)
>  {
> @@ -729,16 +744,53 @@ static int cpu_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
>  	if (!cpumask_test_cpu(smp_processor_id(), &armpmu->supported_cpus))
>  		return NOTIFY_DONE;
>  
> -	/*
> -	 * Always reset the PMU registers on power-up even if
> -	 * there are no events running.
> -	 */
> -	if (cmd == CPU_PM_EXIT && armpmu->reset)
> -		armpmu->reset(armpmu);
> +	if (cmd == CPU_PM_EXIT) {
> +		/*
> +		 * Always reset the PMU registers on power-up even if
> +		 * there are no events running.
> +		 */
> +		if (armpmu->reset)
> +			armpmu->reset(armpmu);
> +
> +		/*
> +		 * Restore the saved pmcr_el0 and pmcntenset_el0
> +		 * if pmc cycle counter is enabled, restore the counter
> +		 */
> +		write_sysreg(get_cpu_var(saved_pmcr_el0), pmcr_el0);
> +		write_sysreg(get_cpu_var(saved_pmcntenset_el0),
> +				pmcntenset_el0);
> +
> +		if (pmc_cycle_counter_enabled()) {
> +			write_sysreg(get_cpu_var(saved_cycle_cnter),
> +					pmccntr_el0);
> +			put_cpu_var(saved_cycle_cnter);
> +		}
> +		put_cpu_var(saved_pmcntenset_el0);
> +		put_cpu_var(saved_pmcr_el0);
> +	}
> +
> +	if (cmd == CPU_PM_ENTER) {
> +		/* If currently pmc cycle counter is enabled,
> +		 * save the counter to percpu section
> +		 */
> +		if (pmc_cycle_counter_enabled()) {
> +			get_cpu_var(saved_cycle_cnter) = read_sysreg(
> +							pmccntr_el0);
> +			put_cpu_var(saved_cycle_cnter);
> +		}
> +
> +		get_cpu_var(saved_pmcr_el0) = read_sysreg(pmcr_el0);
> +
> +		get_cpu_var(saved_pmcntenset_el0) = read_sysreg(
> +							pmcntenset_el0);
> +		put_cpu_var(saved_pmcntenset_el0);
> +		put_cpu_var(saved_pmcr_el0);
> +	}
>  
>  	if (!enabled)
>  		return NOTIFY_OK;
>  
> +	/* if any hw_events is used */
>  	switch (cmd) {
>  	case CPU_PM_ENTER:
>  		armpmu->stop(armpmu);
> @@ -758,7 +810,15 @@ static int cpu_pm_pmu_notify(struct notifier_block *b, unsigned long cmd,
>  
>  static int cpu_pm_pmu_register(struct arm_pmu *cpu_pmu)
>  {
> +	int i;
>  	cpu_pmu->cpu_pm_nb.notifier_call = cpu_pm_pmu_notify;
> +
> +	for_each_possible_cpu(i) {
> +		per_cpu(saved_pmcr_el0, i) = 0;
> +		per_cpu(saved_pmcntenset_el0, i) = 0;
> +		per_cpu(saved_cycle_cnter, i) = 0;
> +	}
> +
>  	return cpu_pm_register_notifier(&cpu_pmu->cpu_pm_nb);
>  }
>  
> -- 
> 2.7.4
>