Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1755633AbaDVKhU (ORCPT ); Tue, 22 Apr 2014 06:37:20 -0400 Received: from cam-admin0.cambridge.arm.com ([217.140.96.50]:39298 "EHLO cam-admin0.cambridge.arm.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754937AbaDVKhO (ORCPT ); Tue, 22 Apr 2014 06:37:14 -0400 Date: Tue, 22 Apr 2014 11:36:42 +0100 From: Will Deacon To: Neil Zhang Cc: "linux@arm.linux.org.uk" , "linux-arm-kernel@lists.infradead.org" , "linux-kernel@vger.kernel.org" , Sudeep Holla , devicetree@vger.kernel.org Subject: Re: [PATCH v4] ARM: perf: save/restore pmu registers in pm notifier Message-ID: <20140422103642.GF7484@arm.com> References: <1398133596-29170-1-git-send-email-zhangwm@marvell.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <1398133596-29170-1-git-send-email-zhangwm@marvell.com> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org Hi Neil, On Tue, Apr 22, 2014 at 03:26:36AM +0100, Neil Zhang wrote: > This adds core support for saving and restoring CPU PMU registers > for suspend/resume support i.e. deeper C-states in cpuidle terms. > This patch adds support only to ARMv7 PMU registers save/restore. > It needs to be extended to xscale and ARMv6 if needed. > > I made this patch because DS-5 is not working on Marvell's CA7 based SoCs. > And it has consulted Sudeep KarkadaNagesha's patch set for multiple PMUs. > > Thanks Will and Sudeep's suggestion to only save / restore used events. Whilst this is a step in the right direction, I'd still like to see the save/restore predicated on something in the device-tree or otherwise. Most SoCs *don't* require these registers to be preserved by software, so we need a way to describe that the PMU is in a power-domain where its state is lost when the CPU goes idle. This doesn't sound like a PMU-specific problem, so there's a possibility that this has been discussed elsewhere, in the context of other IP blocks [adding the devicetree list in case somebody there is aware of any work in this area] Will > diff --git a/arch/arm/include/asm/pmu.h b/arch/arm/include/asm/pmu.h > index ae1919b..3de3db7 100644 > --- a/arch/arm/include/asm/pmu.h > +++ b/arch/arm/include/asm/pmu.h > @@ -83,6 +83,10 @@ struct arm_pmu { > int (*request_irq)(struct arm_pmu *, irq_handler_t handler); > void (*free_irq)(struct arm_pmu *); > int (*map_event)(struct perf_event *event); > + int (*register_pm_notifier)(struct arm_pmu *); > + void (*unregister_pm_notifier)(struct arm_pmu *); > + void (*save_regs)(struct arm_pmu *); > + void (*restore_regs)(struct arm_pmu *); > int num_events; > atomic_t active_events; > struct mutex reserve_mutex; > diff --git a/arch/arm/kernel/perf_event.c b/arch/arm/kernel/perf_event.c > index a6bc431..08822de 100644 > --- a/arch/arm/kernel/perf_event.c > +++ b/arch/arm/kernel/perf_event.c > @@ -326,6 +326,7 @@ static void > armpmu_release_hardware(struct arm_pmu *armpmu) > { > armpmu->free_irq(armpmu); > + armpmu->unregister_pm_notifier(armpmu); > pm_runtime_put_sync(&armpmu->plat_device->dev); > } > > @@ -339,6 +340,7 @@ armpmu_reserve_hardware(struct arm_pmu *armpmu) > return -ENODEV; > > pm_runtime_get_sync(&pmu_device->dev); > + armpmu->register_pm_notifier(armpmu); > err = armpmu->request_irq(armpmu, armpmu_dispatch_irq); > if (err) { > armpmu_release_hardware(armpmu); > diff --git a/arch/arm/kernel/perf_event_cpu.c b/arch/arm/kernel/perf_event_cpu.c > index 51798d7..79e1c06 100644 > --- a/arch/arm/kernel/perf_event_cpu.c > +++ b/arch/arm/kernel/perf_event_cpu.c > @@ -19,6 +19,7 @@ > #define pr_fmt(fmt) "CPU PMU: " fmt > > #include > +#include > #include > #include > #include > @@ -173,6 +174,31 @@ static int cpu_pmu_request_irq(struct arm_pmu *cpu_pmu, irq_handler_t handler) > return 0; > } > > +static int cpu_pmu_pm_notify(struct notifier_block *b, > + unsigned long action, void *v) > +{ > + if (action == CPU_PM_ENTER && cpu_pmu->save_regs) > + cpu_pmu->save_regs(cpu_pmu); > + else if (action == CPU_PM_EXIT && cpu_pmu->restore_regs) > + cpu_pmu->restore_regs(cpu_pmu); > + > + return NOTIFY_OK; > +} > + > +static struct notifier_block cpu_pmu_pm_notifier = { > + .notifier_call = cpu_pmu_pm_notify, > +}; > + > +static int cpu_pmu_register_pm_notifier(struct arm_pmu *cpu_pmu) > +{ > + return cpu_pm_register_notifier(&cpu_pmu_pm_notifier); > +} > + > +static void cpu_pmu_unregister_pm_notifier(struct arm_pmu *cpu_pmu) > +{ > + cpu_pm_unregister_notifier(&cpu_pmu_pm_notifier); > +} > + > static void cpu_pmu_init(struct arm_pmu *cpu_pmu) > { > int cpu; > @@ -187,6 +213,8 @@ static void cpu_pmu_init(struct arm_pmu *cpu_pmu) > cpu_pmu->get_hw_events = cpu_pmu_get_cpu_events; > cpu_pmu->request_irq = cpu_pmu_request_irq; > cpu_pmu->free_irq = cpu_pmu_free_irq; > + cpu_pmu->register_pm_notifier = cpu_pmu_register_pm_notifier; > + cpu_pmu->unregister_pm_notifier = cpu_pmu_unregister_pm_notifier; > > /* Ensure the PMU has sane values out of reset. */ > if (cpu_pmu->reset) > diff --git a/arch/arm/kernel/perf_event_v7.c b/arch/arm/kernel/perf_event_v7.c > index f4ef398..9069310 100644 > --- a/arch/arm/kernel/perf_event_v7.c > +++ b/arch/arm/kernel/perf_event_v7.c > @@ -1237,6 +1237,78 @@ static void armv7_pmnc_dump_regs(struct arm_pmu *cpu_pmu) > } > #endif > > +struct armv7_pmuregs { > + u32 pmc; > + u32 pmcntenset; > + u32 pmintenset; > + u32 pmxevttype[8]; > + u32 pmxevtcnt[8]; > +}; > + > +static DEFINE_PER_CPU(struct armv7_pmuregs, pmu_regs); > + > +static void armv7pmu_reset(void *info); > + > +static void armv7pmu_save_regs(struct arm_pmu *cpu_pmu) > +{ > + struct pmu_hw_events *events = cpu_pmu->get_hw_events(); > + struct armv7_pmuregs *regs; > + int bit; > + > + /* Check whether there are events used */ > + bit = find_first_bit(events->used_mask, cpu_pmu->num_events); > + if (bit >= cpu_pmu->num_events) > + return; > + > + regs = this_cpu_ptr(&pmu_regs); > + for_each_set_bit(bit, events->used_mask, cpu_pmu->num_events) { > + if (bit) { > + armv7_pmnc_select_counter(bit); > + asm volatile("mrc p15, 0, %0, c9, c13, 1" > + : "=r"(regs->pmxevttype[bit])); > + asm volatile("mrc p15, 0, %0, c9, c13, 2" > + : "=r"(regs->pmxevtcnt[bit])); > + } else > + asm volatile("mrc p15, 0, %0, c9, c13, 0" > + : "=r" (regs->pmxevtcnt[0])); > + } > + > + asm volatile("mrc p15, 0, %0, c9, c12, 1" : "=r" (regs->pmcntenset)); > + asm volatile("mrc p15, 0, %0, c9, c14, 1" : "=r" (regs->pmintenset)); > + asm volatile("mrc p15, 0, %0, c9, c12, 0" : "=r" (regs->pmc)); > +} > + > +static void armv7pmu_restore_regs(struct arm_pmu *cpu_pmu) > +{ > + struct pmu_hw_events *events = cpu_pmu->get_hw_events(); > + struct armv7_pmuregs *regs; > + int bit; > + > + /* Check whether there are events used */ > + bit = find_first_bit(events->used_mask, cpu_pmu->num_events); > + if (bit >= cpu_pmu->num_events) > + return; > + > + armv7pmu_reset(cpu_pmu); > + > + regs = this_cpu_ptr(&pmu_regs); > + for_each_set_bit(bit, events->used_mask, cpu_pmu->num_events) { > + if (bit) { > + armv7_pmnc_select_counter(bit); > + asm volatile("mcr p15, 0, %0, c9, c13, 1" > + : : "r"(regs->pmxevttype[bit])); > + asm volatile("mcr p15, 0, %0, c9, c13, 2" > + : : "r"(regs->pmxevtcnt[bit])); > + } else > + asm volatile("mcr p15, 0, %0, c9, c13, 0" > + : : "r" (regs->pmxevtcnt[0])); > + } > + > + asm volatile("mcr p15, 0, %0, c9, c12, 1" : : "r" (regs->pmcntenset)); > + asm volatile("mcr p15, 0, %0, c9, c14, 1" : : "r" (regs->pmintenset)); > + asm volatile("mcr p15, 0, %0, c9, c12, 0" : : "r" (regs->pmc)); > +} > + > static void armv7pmu_enable_event(struct perf_event *event) > { > unsigned long flags; > @@ -1528,6 +1600,8 @@ static void armv7pmu_init(struct arm_pmu *cpu_pmu) > cpu_pmu->start = armv7pmu_start; > cpu_pmu->stop = armv7pmu_stop; > cpu_pmu->reset = armv7pmu_reset; > + cpu_pmu->save_regs = armv7pmu_save_regs; > + cpu_pmu->restore_regs = armv7pmu_restore_regs; > cpu_pmu->max_period = (1LLU << 32) - 1; > }; > > -- > 1.7.9.5 > > -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/