linux-kernel - Re: [PATCH v6 6/7] arm_pmu: Introduce pmu_irq

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] [day] [month] [year] [list]

Message-ID: <48af8175-3908-8de3-783f-c61522e9c6fc@arm.com>
Date:   Wed, 23 Sep 2020 16:46:00 +0100
From:   Alexandru Elisei <alexandru.elisei@....com>
To:     Will Deacon <will@...nel.org>
Cc:     linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        mark.rutland@....com, maz@...nel.org, catalin.marinas@....com,
        swboyd@...omium.org, sumit.garg@...aro.org,
        Julien Thierry <julien.thierry@....com>,
        Julien Thierry <julien.thierry.kdev@...il.com>,
        Will Deacon <will.deacon@....com>
Subject: Re: [PATCH v6 6/7] arm_pmu: Introduce pmu_irq_ops

Hi Will,

On 9/21/20 2:55 PM, Will Deacon wrote:
> On Wed, Aug 19, 2020 at 02:34:18PM +0100, Alexandru Elisei wrote:
>> From: Julien Thierry <julien.thierry@....com>
>>
>> Currently the PMU interrupt can either be a normal irq or a percpu irq.
>> Supporting NMI will introduce two cases for each existing one. It becomes
>> a mess of 'if's when managing the interrupt.
>>
>> Define sets of callbacks for operations commonly done on the interrupt. The
>> appropriate set of callbacks is selected at interrupt request time and
>> simplifies interrupt enabling/disabling and freeing.
>>
>> Cc: Julien Thierry <julien.thierry.kdev@...il.com>
>> Cc: Will Deacon <will.deacon@....com>
>> Cc: Mark Rutland <mark.rutland@....com>
>> Signed-off-by: Julien Thierry <julien.thierry@....com>
>> Signed-off-by: Alexandru Elisei <alexandru.elisei@....com>
>> ---
>>  drivers/perf/arm_pmu.c | 86 ++++++++++++++++++++++++++++++++++--------
>>  1 file changed, 70 insertions(+), 16 deletions(-)
>>
>> diff --git a/drivers/perf/arm_pmu.c b/drivers/perf/arm_pmu.c
>> index df352b334ea7..17e5952d21e4 100644
>> --- a/drivers/perf/arm_pmu.c
>> +++ b/drivers/perf/arm_pmu.c
>> @@ -26,8 +26,46 @@
>>  
>>  #include <asm/irq_regs.h>
>>  
>> +static int armpmu_count_irq_users(const int irq);
>> +
>> +struct pmu_irq_ops {
>> +	void (*enable_pmuirq)(unsigned int irq);
>> +	void (*disable_pmuirq)(unsigned int irq);
>> +	void (*free_pmuirq)(unsigned int irq, int cpu, void __percpu *devid);
>> +};
>> +
>> +static void armpmu_free_pmuirq(unsigned int irq, int cpu, void __percpu *devid)
>> +{
>> +	free_irq(irq, per_cpu_ptr(devid, cpu));
>> +}
>> +
>> +static const struct pmu_irq_ops pmuirq_ops = {
>> +	.enable_pmuirq = enable_irq,
>> +	.disable_pmuirq = disable_irq_nosync,
>> +	.free_pmuirq = armpmu_free_pmuirq
>> +};
>> +
>> +static void armpmu_enable_percpu_pmuirq(unsigned int irq)
>> +{
>> +	enable_percpu_irq(irq, IRQ_TYPE_NONE);
>> +}
>> +
>> +static void armpmu_free_percpu_pmuirq(unsigned int irq, int cpu,
>> +				   void __percpu *devid)
>> +{
>> +	if (armpmu_count_irq_users(irq) == 1)
>> +		free_percpu_irq(irq, devid);
>> +}
>> +
>> +static const struct pmu_irq_ops percpu_pmuirq_ops = {
>> +	.enable_pmuirq = armpmu_enable_percpu_pmuirq,
>> +	.disable_pmuirq = disable_percpu_irq,
>> +	.free_pmuirq = armpmu_free_percpu_pmuirq
>> +};
>> +
>>  static DEFINE_PER_CPU(struct arm_pmu *, cpu_armpmu);
>>  static DEFINE_PER_CPU(int, cpu_irq);
>> +static DEFINE_PER_CPU(const struct pmu_irq_ops *, cpu_irq_ops);
> Would it make sense to put this in a structure alongside the irq?
It doesn't really work, because we need the irq number to be percpu for
armpmu_free_irq() to work correctly. If we have a percpu pointer to a struct, the
first cpu that frees the irq will set it to 0, and all subsequent CPUs that share
the same struct will read 0 as the irq number, which will trigger the WARN_ON and
then return early.
>
>>  
>>  static inline u64 arm_pmu_event_max_period(struct perf_event *event)
>>  {
>> @@ -544,6 +582,19 @@ static int armpmu_count_irq_users(const int irq)
>>  	return count;
>>  }
>>  
>> +static const struct pmu_irq_ops *armpmu_find_irq_ops(int irq)
>> +{
>> +	int cpu;
>> +
>> +	for_each_possible_cpu(cpu) {
>> +		if (per_cpu(cpu_irq, cpu) == irq
>> +		    && per_cpu(cpu_irq_ops, cpu))
>> +			return per_cpu(cpu_irq_ops, cpu);
>> +	}
> nit, but you could make this a bit more readable:
>
> 	struct pmu_irq_ops *ops = NULL;
>
> 	for_each_possible_cpu(cpu) {
> 		if (per_cpu(cpu_irq, cpu) != irq)
> 			continue;
>
> 		ops = per_cpu(cpu_irq_ops, cpu);
> 		if (ops)
> 			break;
> 	}
>
> 	return ops;

That looks better, I will change it.

Thanks,
Alex