x86 / CPU: Avoid unnecessary IPIs in arch_freq_get_on_cpu()
authorRafael J. Wysocki <rafael.j.wysocki@intel.com>
Mon, 13 Nov 2017 01:15:39 +0000 (02:15 +0100)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 Nov 2017 03:42:39 +0000 (19:42 -0800)
Even though aperfmperf_snapshot_khz() caches the samples.khz value to
return if called again in a sufficiently short time, its caller,
arch_freq_get_on_cpu(), still uses smp_call_function_single() to run it
which may allow user space to trigger an IPI storm by reading from the
scaling_cur_freq cpufreq sysfs file in a tight loop.

To avoid that, move the decision on whether or not to return the cached
samples.khz value to arch_freq_get_on_cpu().

This change was part of commit 941f5f0f6ef5 ("x86: CPU: Fix up "cpu MHz"
in /proc/cpuinfo"), but it was not the reason for the revert and it
remains applicable.

Fixes: 4815d3c56d1e (cpufreq: x86: Make scaling_cur_freq behave more as expected)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: WANG Chao <chao.wang@ucloud.cn>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/x86/kernel/cpu/aperfmperf.c

index 0ee83321a3136fcca7a00a3b7e6c375e7a51e13f..957813e0180d278563bf33289c54c7462ca83a8a 100644 (file)
@@ -42,10 +42,6 @@ static void aperfmperf_snapshot_khz(void *dummy)
        s64 time_delta = ktime_ms_delta(now, s->time);
        unsigned long flags;
 
        s64 time_delta = ktime_ms_delta(now, s->time);
        unsigned long flags;
 
-       /* Don't bother re-computing within the cache threshold time. */
-       if (time_delta < APERFMPERF_CACHE_THRESHOLD_MS)
-               return;
-
        local_irq_save(flags);
        rdmsrl(MSR_IA32_APERF, aperf);
        rdmsrl(MSR_IA32_MPERF, mperf);
        local_irq_save(flags);
        rdmsrl(MSR_IA32_APERF, aperf);
        rdmsrl(MSR_IA32_MPERF, mperf);
@@ -74,6 +70,7 @@ static void aperfmperf_snapshot_khz(void *dummy)
 
 unsigned int arch_freq_get_on_cpu(int cpu)
 {
 
 unsigned int arch_freq_get_on_cpu(int cpu)
 {
+       s64 time_delta;
        unsigned int khz;
 
        if (!cpu_khz)
        unsigned int khz;
 
        if (!cpu_khz)
@@ -82,6 +79,12 @@ unsigned int arch_freq_get_on_cpu(int cpu)
        if (!static_cpu_has(X86_FEATURE_APERFMPERF))
                return 0;
 
        if (!static_cpu_has(X86_FEATURE_APERFMPERF))
                return 0;
 
+       /* Don't bother re-computing within the cache threshold time. */
+       time_delta = ktime_ms_delta(ktime_get(), per_cpu(samples.time, cpu));
+       khz = per_cpu(samples.khz, cpu);
+       if (khz && time_delta < APERFMPERF_CACHE_THRESHOLD_MS)
+               return khz;
+
        smp_call_function_single(cpu, aperfmperf_snapshot_khz, NULL, 1);
        khz = per_cpu(samples.khz, cpu);
        if (khz)
        smp_call_function_single(cpu, aperfmperf_snapshot_khz, NULL, 1);
        khz = per_cpu(samples.khz, cpu);
        if (khz)