| 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
| ||
|
Message-ID: <20160229180048.GB20610@hr-amur2>
Date: Tue, 1 Mar 2016 02:00:50 +0800
From: Huang Rui <ray.huang@....com>
To: Thomas Gleixner <tglx@...utronix.de>
CC: Borislav Petkov <bp@...e.de>,
Peter Zijlstra <peterz@...radead.org>,
"Ingo Molnar" <mingo@...nel.org>,
Andy Lutomirski <luto@...capital.net>,
"Robert Richter" <rric@...nel.org>,
Jacob Shin <jacob.w.shin@...il.com>,
"Arnaldo Carvalho de Melo" <acme@...nel.org>,
Kan Liang <kan.liang@...el.com>,
<linux-kernel@...r.kernel.org>, <spg_linux_kernel@....com>,
<x86@...nel.org>,
Suravee Suthikulpanit <suravee.suthikulpanit@....com>,
Aravind Gopalakrishnan <Aravind.Gopalakrishnan@....com>,
Borislav Petkov <bp@...en8.de>,
"Fengguang Wu" <fengguang.wu@...el.com>,
Guenter Roeck <linux@...ck-us.net>
Subject: Re: [PATCH v5] perf/x86/amd/power: Add AMD accumulated power
reporting mechanism
Hi Thomas,
Thank you to your comments. :-)
On Fri, Feb 26, 2016 at 11:18:28AM +0100, Thomas Gleixner wrote:
> On Fri, 26 Feb 2016, Huang Rui wrote:
>
> > +
> > +static DEFINE_PER_CPU(struct power_pmu_masks *, amd_power_pmu);
>
> This is complete overkill, really.
>
> > +static int power_cpu_exit(int cpu)
> > +{
> > + struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
> > + int target = nr_cpumask_bits;
> > + int ret = 0;
> > +
> > + cpumask_copy(pmu->mask, topology_sibling_cpumask(cpu));
> > +
> > + cpumask_clear_cpu(cpu, &cpu_mask);
> > + cpumask_clear_cpu(cpu, pmu->mask);
> > +
> > + if (!cpumask_and(pmu->tmp_mask, pmu->mask, cpu_online_mask))
> > + goto out;
> > + /*
> > + * Find a new CPU on the same compute unit, if was set in cpumask
> > + * and still some CPUs on compute unit. Then move on to the new CPU.
> > + */
> > + target = cpumask_any(pmu->tmp_mask);
> > + if (target < nr_cpumask_bits && target != cpu)
> > + cpumask_set_cpu(target, &cpu_mask);
> > +
> > + WARN_ON(cpumask_empty(&cpu_mask));
> > +
> > +out:
> > + /*
> > + * Migrate event and context to new CPU.
> > + */
> > + if (target < nr_cpumask_bits)
> > + perf_pmu_migrate_context(&pmu_class, cpu, target);
> > +
> > + return ret;
>
> Sigh. That whole thing can be simply done with:
>
> if (!cpumask_test_and_clear_cpu(cpu, &cpu_mask))
> return;
>
> target = cpumask_any_but(topology_sibling_cpumask(cpu), cpu);
> if (target < nr_cpumask_bits) {
> cpumask_set_cpu(target, &uncore_cpu_mask);
> perf_pmu_migrate_context(&pmu_class, cpu, target);
> }
>
>
> > +
> > +}
> > +
> > +static int power_cpu_init(int cpu)
> > +{
> > + struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
> > +
> > + if (!pmu)
> > + return 0;
> > +
> > + if (!cpumask_and(pmu->mask, topology_sibling_cpumask(cpu), &cpu_mask))
> > + cpumask_set_cpu(cpu, &cpu_mask);
>
> And that's equally convoluted and can be done with:
>
> target = cpumask_any_and(&cpu_mask, topology_sibling_cpumask(cpu));
> if (target < nr_cpu_ids)
> cpumask_set_cpu(cpu, &cpu_mask);
>
> > +static int power_cpu_prepare(int cpu)
> > +{
> > + struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
> > + int phys_id = topology_physical_package_id(cpu);
> > + int ret = 0;
> > +
> > + if (pmu)
> > + return 0;
> > +
> > + if (phys_id < 0)
> > + return -EINVAL;
> > +
> > + pmu = kzalloc_node(sizeof(*pmu), GFP_KERNEL, cpu_to_node(cpu));
> > + if (!pmu)
> > + return -ENOMEM;
> > +
> > + if (!zalloc_cpumask_var(&pmu->mask, GFP_KERNEL)) {
> > + ret = -ENOMEM;
> > + goto out;
> > + }
> > +
> > + if (!zalloc_cpumask_var(&pmu->tmp_mask, GFP_KERNEL)) {
> > + ret = -ENOMEM;
> > + goto out1;
> > + }
> > +
> > + per_cpu(amd_power_pmu, cpu) = pmu;
> > + return 0;
> > +
> > +out1:
> > + free_cpumask_var(pmu->mask);
> > +out:
> > + kfree(pmu);
> > +
> > + return ret;
> > +}
> > +
> > +static void power_cpu_kfree(int cpu)
> > +{
> > + struct power_pmu_masks *pmu = per_cpu(amd_power_pmu, cpu);
> > +
> > + if (!pmu)
> > + return;
> > +
> > + free_cpumask_var(pmu->mask);
> > + free_cpumask_var(pmu->tmp_mask);
> > + kfree(pmu);
> > +
> > + per_cpu(amd_power_pmu, cpu) = NULL;
> > +}
>
> Both prepare and free can go away.
>
OK, I got your point. After enhounce the implementation of cpu
notifier part of this driver, we can avoid to use weird
"DEFINE_PER_CPU(struct power_pmu_masks *, amd_power_pmu)". Thanks.
> > +static int __init amd_power_pmu_init(void)
> > +{
> > + int i, ret;
> > + u64 tmp;
> > +
> > + if (!x86_match_cpu(cpu_match))
> > + return 0;
> > +
> > + if (!boot_cpu_has(X86_FEATURE_ACC_POWER))
> > + return -ENODEV;
> > +
> > + cores_per_cu = amd_get_cores_per_cu();
> > + cu_num = boot_cpu_data.x86_max_cores / cores_per_cu;
>
> Please use the new package management functions which are on the way to tip.
>
Can you give me some hints?
> > +
> > + if (WARN_ON_ONCE(cu_num > MAX_CUS))
> > + return -EINVAL;
> > +
> > + cpu_pwr_sample_ratio = cpuid_ecx(0x80000007);
> > +
> > + if (rdmsrl_safe(MSR_F15H_CU_MAX_PWR_ACCUMULATOR, &tmp)) {
> > + pr_err("Failed to read max compute unit power accumulator MSR\n");
> > + return -ENODEV;
> > + }
> > + max_cu_acc_power = tmp;
> > +
> > + cpu_notifier_register_begin();
> > +
> > + /* Choose one online core of each compute unit. */
> > + for (i = 0; i < boot_cpu_data.x86_max_cores; i += cores_per_cu) {
> > + WARN_ON(cpumask_empty(topology_sibling_cpumask(i)));
> > + cpumask_set_cpu(cpumask_any(topology_sibling_cpumask(i)), &cpu_mask);
> > + }
>
>
>
> > +
> > + for_each_present_cpu(i) {
> > + ret = power_cpu_prepare(i);
> > + if (ret) {
> > + /* Unwind on [0 ... i-1] CPUs. */
> > + while (i--)
> > + power_cpu_kfree(i);
> > + goto out;
> > + }
>
> Not needed.
>
> > + ret = power_cpu_init(i);
> > + if (ret) {
> > + /* Unwind on [0 ... i] CPUs. */
> > + while (i >= 0)
> > + power_cpu_kfree(i--);
> > + goto out;
> > + }
>
> You cannot issue the CPU STARTING callback on present, but not online cpus.
>
Do you mean we should change for_each_present_cpu to
for_each_online_cpu?
My orignal intent here, it's to allocate data structures of
"power_pmu_masks" for each core. But now, I think we needn't below
codes, right?
for_each_online_cpu(i)
power_cpu_init(i);
> > + }
> > +
> > + __perf_cpu_notifier(power_cpu_notifier);
> > +
> > + ret = perf_pmu_register(&pmu_class, "power", -1);
> > + if (WARN_ON(ret)) {
> > + pr_warn("AMD Power PMU registration failed\n");
>
>
> So that leaves the notifier installed and leaks all the allocated memory.
>
OK, do you mean "issue CPU_STARTING callback on present cpus" will
cause the memory leak here? Could you please explain more?
> > + goto out;
> > + }
> > +
> > + pr_info("AMD Power PMU detected, %d compute units\n", cu_num);
> > +
> > +out:
> > + cpu_notifier_register_done();
> > +
> > + return ret;
> > +}
> > +device_initcall(amd_power_pmu_init);
>
> Please make this as a module right away. There is no reason to compile all
> this in.
>
OK, will update.
Thanks,
Rui
Powered by blists - more mailing lists